PVT — from the sample to the number you trust

● Reservoir Fluids · Sep 9, 2026 · 24 min read

Almost every volume a reservoir engineer reports passes through a PVT report on its way out. OOIP divides by B_oi. Material balance stands on B_o, R_s and B_g. Nodal analysis needs viscosity and solution gas at every node. Yet the PVT report is the input least often questioned — it arrives bound, tabulated and authoritative, and gets typed straight into the model. This guide is about what is actually inside it, which parts can be wrong, and how to check.

The stakes are arithmetic and unforgiving. Original oil in place is inversely proportional to B_oi, so a formation volume factor 5% high makes OOIP 4.8% low — 1/1.05 of the true value — without any modelling error being involved. No amount of seismic, petrophysics or simulation effort recovers that. The single number sits at the front of the calculation and scales everything behind it.

A PVT report is not data. It is the result of an experiment on a sample that may or may not be the reservoir fluid, processed through conventions that may or may not match how you intend to use it.

First decide what fluid you have^[1][2]

Every subsequent choice — how to sample, which laboratory study to request, whether a correlation is admissible, whether you need an equation of state — follows from the fluid type. Get this wrong and the rest of the programme is aimed at the wrong target.

The classification is geometric. Draw the phase envelope in pressure–temperature space, mark the critical point and the cricondentherm, then locate reservoir temperature relative to them. That single comparison separates the five types.

Figure 1The classification is one comparison. If reservoir temperature is below the critical temperature T₊, the fluid is an oil and the reservoir pressure path crosses a bubble-point line. If it lies between T₊ and the cricondentherm, the fluid is a gas condensate and the path crosses a dew-point line, dropping liquid out inside the reservoir as pressure falls. Above the cricondentherm no liquid ever forms in the reservoir, though a wet gas still drops liquid in the separator. The nearer reservoir temperature sits to the critical point, the more violently properties change with a small pressure step — which is why volatile oils and rich condensates are the two types where correlations stop being defensible.

Type	Initial GOR (scf/STB)	Stock-tank gravity	What breaks if misclassified
Dry gas	no liquid	—	Little; the simple case
Wet gas	> 15,000	> 50°API	Condensate volumes omitted from the revenue forecast
Gas condensate	3,200 – 15,000	40 – 60°API	Liquid dropout and the resulting near-well gas relative permeability loss are both missed
Volatile oil	2,000 – 3,200	> 40°API	Black-oil treatment under-predicts recovery; a large part of stock-tank liquid arrives from the gas phase
Black oil	< 2,000	< 45°API	Little; the case every correlation was built for

The boundaries are indicative, not definitions

The GOR and gravity ranges above are field rules of thumb, and different texts quote slightly different numbers.^[1][2][3] The definition is the position of reservoir temperature relative to the critical point, which only a compositional analysis and a phase-behaviour calculation can establish. Treat a GOR near a boundary as a reason to look at the composition, not as a classification. A fluid at 2,800 scf/STB may be a volatile oil or a very rich condensate, and the two require different laboratory studies.

Sampling: where most PVT error is created^[4]

A laboratory can measure a sample to a fraction of a percent. None of that precision survives if the sample is not the reservoir fluid. In practice, sampling contributes far more error to a PVT dataset than the measurements do, and unlike measurement error it is not random — it is nearly always biased in one direction.

The mechanism of the bias

If flowing bottomhole pressure has fallen below the saturation pressure, gas has already come out of solution in the near-well region. Gas and oil then flow at different rates, because relative permeability treats them differently. Whatever is captured at that point is a mixture in proportions set by the flow, not by the reservoir. The near-well region has become a separator that nobody calibrated.

The direction of the bias is not fixed, and it is worth being careful here because both directions occur and only one of them is detectable afterwards.

Below the critical gas saturation, the liberated gas is trapped and immobile. Only oil flows, and that oil has already given up gas to the stationary phase — so the sample is gas-deficient, and reports a saturation pressure lower than the true value. Once gas saturation exceeds the critical value the gas becomes mobile and, being far less viscous than the oil, flows preferentially. The sample is then gas-enriched, and reports a saturation pressure higher than the true value.

A single sample cannot tell you which regime produced it. What it can tell you is when it is definitely wrong: a bubble point reported above measured reservoir pressure is not possible for a fluid that was single-phase in place, and that is the most frequently caught error in a PVT review. The low-bubble-point case is the more dangerous of the two, precisely because nothing inside the report contradicts it — it has to be caught at the sampling conditions or not at all.

Figure 2The sampling decision reduces to one physical question: is the fluid single-phase where the sample is taken? Bottomhole sampling is preferred when it is, because it avoids the recombination step entirely. When it is not, and rate reduction cannot restore single-phase flow, surface sampling with recombination is the correct answer rather than a fallback. For gas condensates the rule inverts: separator sampling is standard practice because a bottomhole sample can pick up accumulated near-well condensate that is not representative of the flowing stream.

Well conditioning

Conditioning means producing the well at a reduced, stable rate for long enough that the gas-depleted near-well fluid is drawn back into the wellbore and replaced by fluid at original composition. The measure of success is a stable producing GOR, not elapsed time. A well whose GOR is still drifting has not finished conditioning, whatever the programme said. Recognised sampling practice for petroleum reservoir fluids is set out in API Recommended Practice 44.^[4]

Recombination

Surface sampling collects separator liquid and separator gas in two containers and recombines them in the laboratory at the producing gas–oil ratio measured while the samples were taken. That measured GOR is the weakest link in the whole procedure: it appears directly in the recombination ratio, so an error in it propagates straight into every recombined property, and unlike a laboratory measurement it is taken under field conditions with field metering.

Two conditions must hold and are worth checking on the field data sheet before the samples ever leave. The separator must be stable — pressure, temperature and levels steady — for the full sampling period. And the GOR must be referenced to the same separator stage the liquid sample came from, at the recorded separator pressure and temperature, not to a field-standard GOR from a different stage.

Oil-based mud contamination

A sample from a well drilled with oil-based mud can contain mud filtrate, and the base oil is heavy relative to the reservoir fluid. Contamination therefore lowers the apparent GOR and the apparent saturation pressure, and raises the apparent stock-tank density and viscosity — the opposite direction to the flow-induced bias described above, which is why the two can partially mask each other.

The laboratory can quantify contamination from the composition, typically by identifying the base-oil signature in the heavier fractions, and can numerically remove it. What matters for the reader of the report is that this correction was made, by how much, and whether the corrected composition was then used for everything downstream. A report that mentions contamination in the narrative but tabulates uncorrected properties is a common and expensive trap.

What the laboratory actually does^[1][3]

A PVT report is a bundle of separate experiments, each answering a different question. They are not interchangeable, and the most common misuse of a PVT report is taking a number from the wrong experiment.

Study	Procedure	What it gives you	Required for
Constant composition expansion (CCE, flash liberation)	Cell pressure lowered in steps; nothing is removed, so total composition is constant throughout	Saturation pressure, relative volume, undersaturated compressibility, liquid dropout curve for condensates	Every fluid
Differential liberation (DL)	Below P_b, liberated gas is removed at each step at reservoir temperature; composition of the remaining liquid changes continuously	B_oD, R_sD, gas gravity and Z per stage, residual oil density	Oils
Separator tests	Sample flashed through one or more separator stages to stock tank, at chosen P and T	B_ofb, R_sfb, shrinkage, stock-tank gravity, optimum separator conditions	Oils — and required to convert the DL data
Constant volume depletion (CVD)	Gas removed at each step so that cell volume returns to the original, mimicking a reservoir where liquid dropout stays behind	Produced-stream composition per stage, retrograde liquid saturation, two-phase Z	Gas condensates, volatile oils
Swelling / multi-contact	Injection gas or solvent added in increments; saturation pressure and swollen volume measured	Swelling factor, minimum miscibility pressure behaviour	Gas injection and miscible EOR studies
Viscosity	Rolling-ball or capillary viscometer at reservoir temperature over the pressure range	μ_o above and below P_b, μ_g	Anything involving flow

The distinction that matters most is between the two liberation processes. Differential liberation removes gas as it forms, at reservoir temperature — a reasonable analogue for what happens in the reservoir, where liberated gas becomes mobile and leaves the oil behind. Separator flash keeps the phases in contact down to stock-tank conditions — a reasonable analogue for what happens in the tubing and surface facilities. Real production does both, in sequence.

The conversion nobody checks^[1][3]

This is the single most common error in the practical use of a PVT report, and it is worth being precise about because it is entirely avoidable.

The differential liberation table reports B_oD and R_sD referenced to residual oil at 60°F — the volume left in the cell at the end of the experiment. Every engineering calculation you will do, however, is referenced to stock-tank oil through the actual separator train. Those are different reference volumes. Reading B_oD straight out of the DL table and calling it B_o introduces an error of the order of several percent, in the direction that overstates B_o and therefore understates OOIP.

Converting differential data to a field basis B_o = B_oD × ( B_ofb / B_oDb )

R_s = R_sfb − ( R_sDb − R_sD ) × ( B_ofb / B_oDb )

where:
  B_oD, R_sD = differential values at the pressure of interest
  B_oDb, R_sDb = differential values at the bubble point
  B_ofb, R_sfb = separator-test values at the bubble point,
    for the separator train you actually intend to operate

Three things follow from the form of these expressions, and each is a check you can perform on any report in under a minute.

The correction is a single constant ratio. B_ofb/B_oDb does not vary with pressure. If a converted table shows the ratio between the differential and field-basis B_o drifting with pressure, the conversion was not done this way.
At the bubble point the conversion must reproduce the separator test exactly. Substituting B_oD = B_oDb gives B_o = B_ofb, and R_sD = R_sDb gives R_s = R_sfb. Any converted table failing this identity at P_b has an arithmetic error, and it is the first line to check.
The conversion is separator-specific. Different separator pressures produce different B_ofb and therefore a different converted table from identical laboratory data. A PVT dataset is only valid for the facilities configuration it was converted for. If the development concept changes the separator train, the conversion has to be redone.

Figure 3The two curves have the same shape because the conversion is a single multiplicative constant. The differential curve sits above the field basis over the whole range, so using it unconverted inflates B_o and deflates OOIP by the same proportion. Both peak at the bubble point — above P_b the oil is single-phase and merely compressed, so B_o falls as pressure rises; below P_b gas leaves solution and the remaining oil shrinks. The gap at P_b is the entire correction, and it is set by the separator train.

Checking the report for internal consistency^[5][6]

Laboratories make errors, samples degrade in transit, and transcription happens. Three checks catch most of what goes wrong, and none of them requires anything beyond the report itself. They are checks of internal consistency: they cannot prove a sample was representative, only that the measurements on it hang together.

The Y-function — smoothness of the CCE data below P_b

Y-function Y = ( P_sat − P ) / [ P × ( V_rel − 1 ) ]

where V_rel = V / V_sat is the relative volume from the CCE,
evaluated at pressures below the saturation pressure

Plotted against pressure, Y should fall on a straight or very gently curved line. The function is a rearrangement of the relative-volume data that amplifies scatter, so points that look acceptable on a V_rel plot become visibly wrong on a Y-function plot. A single point off the line is a suspect measurement. A systematic curvature or a break in slope points at something worse — commonly an incorrect saturation pressure, because P_sat appears in the numerator and an error in it bends the whole line.

Figure 4The Y-function turns a smooth-looking relative-volume table into a straight-line test. The outlier here is a single cell reading that looks unremarkable in the original tabulation. What the plot cannot tell you is whether the sample was representative — a perfectly conditioned experiment on the wrong fluid gives a perfectly straight line.

Equilibrium ratios — the Hoffmann–Crump–Hocott plot

Where stage-by-stage compositions are reported — separator tests, and every CVD stage — equilibrium ratios K_i = y_i/x_i can be computed per component and plotted as log(K_iP) against a characterisation factor F_i built from each component's boiling point and critical properties.^[5] For a consistent set of compositions the points fall close to a straight line for each stage. Components that sit off the line, or a line whose slope changes erratically between stages, indicate a compositional analysis problem rather than a physical one.

Figure 5Each depletion stage should produce its own straight line, with the lines shifting in an orderly way as pressure falls. Disorderly shifts, or a component persistently off trend, point at the compositional analysis. This diagnostic is also what makes a dataset usable for equation-of-state tuning: an EOS fitted to inconsistent K-values will reproduce the inconsistency and call it a match.

Material balance on the laboratory data itself

The differential liberation is a closed experiment: the mass that goes in must come out. Sum the liberated gas across all stages, add the residual oil, and compare against the original cell charge. Compositional laboratories perform this check routinely; it is not always printed. Asking for it is reasonable, and a laboratory that cannot produce it has told you something.

The check these cannot perform

All three tests examine whether the report is internally coherent. None can tell you whether the sample represented the reservoir. That question is answered by the sampling conditions — the flowing bottomhole pressure at the time of sampling, the stability of the producing GOR, and whether the reported saturation pressure is physically admissible against the measured reservoir pressure. It is answered on the field data sheet, not in the laboratory.

Correlations — legitimate use and misuse^[7][8][9]

When there is no PVT report, correlations fill the gap. They are regressions fitted to sets of laboratory measurements, which means each one is an interpolation inside the dataset that produced it. Used inside that range they are defensible engineering. Used outside it they are extrapolation dressed as measurement, and the error is not estimable from the correlation.

Property	Correlation	Development basis	Notes
P_b, R_s, B_o	Standing (1947)	California crudes	The original; still reasonable for black oils of similar character
R_s, B_o, c_o, μ_o	Vazquez & Beggs (1980)	Wide multi-source dataset, split above and below 30°API	Requires gas gravity corrected to a 100 psig separator reference — a step routinely skipped
P_b, B_o	Glasø (1980)^[10]	North Sea crudes	Preferred where the fluid resembles that province
P_b, B_o	Al-Marhoun (1988)^[11]	Middle East crudes	Same logic, different province
μ_o dead / saturated	Beggs & Robinson (1975)	Broad dataset	Dead-oil viscosity is the least reliable property in all of PVT correlation — errors of tens of percent are ordinary
Z-factor	Standing & Katz chart^[12]; Dranchuk & Abou-Kassem fit^[13]	Natural gases, sweet	Correct for H₂S and CO₂ before use; sour gas without correction is a real and large error

Three rules make correlation use defensible rather than merely convenient.

Match the province and the fluid character, then check the range. API gravity, gas gravity, temperature and pressure should all sit inside the correlation's stated development envelope. A correlation built on 30–45°API crudes says nothing trustworthy about a 16°API oil.
If any laboratory data exists, anchor to it. One measured bubble point is enough to shift a correlation onto the right level, and a correlation anchored to a single measurement is far better than either the raw correlation or the single point alone.
Do not use black-oil correlations on volatile oils or condensates. They were not built for fluids where a large fraction of the stock-tank liquid arrives from the gas phase, and the black-oil formulation itself — one gas dissolving in one oil — stops describing the physics.

When you need an equation of state^[2][14][15]

Black-oil tables carry two parameters between the phases: how much gas dissolves in the oil, and how the volumes change. That is sufficient when composition is effectively fixed. It stops being sufficient when composition itself changes with pressure or position — which is exactly the case for gas condensates, volatile oils, gas injection and miscible EOR, and any reservoir with a compositional gradient over its column.

For those, a cubic equation of state is used to compute phase behaviour from composition. The two in general use are Soave–Redlich–Kwong^[16] and Peng–Robinson; the latter was developed specifically to improve liquid-density prediction over SRK, which is why it predominates in reservoir work.^[14]

Peng–Robinson equation of state P = RT / (v − b) − aα / [ v² + 2bv − b² ]

a = 0.45724 R²T_c² / P_c · b = 0.07780 R T_c / P_c

α = [ 1 + m(1 − √(T/T_c)) ]²
m = 0.37464 + 1.54226ω − 0.26992ω²

ω = acentric factor

Tuning, and the way it goes wrong

An untuned EOS will not reproduce a measured PVT dataset. Tuning adjusts a small number of parameters — usually the critical properties and acentric factor of the lumped heavy fraction, plus selected binary interaction coefficients — until computed behaviour matches the laboratory.

The failure mode is over-fitting, and it is seductive because it looks like success. Enough free parameters will match any dataset, including its errors. The model then reproduces the experiments beautifully and predicts badly outside them, which is precisely the region a compositional model exists to explore.

Three disciplines that keep a tuned EOS honest

Adjust characterisation parameters, not measured ones. The heavy fraction's properties are estimated and are legitimate targets. Measured compositions of light components are data, and moving them to improve a match is fitting away the evidence.

Hold something back. Tune on the CCE and DL; check against the separator tests or a later CVD stage that was not used in the fit. An EOS that matches only what it was tuned on has been fitted, not validated — the same blind-test discipline that applies to any other model on measured data.

Keep the parameters physical. A tuned critical temperature far outside the plausible range for that fraction is a warning, even when the match improves. It usually means the model is absorbing an inconsistency in the data — which is why the consistency checks above come first.

What to hand downstream

The last step is delivery, and it is where a correct dataset can still be used wrongly.

Consumer	Needs	The usual mistake
Oil material balance	Field-basis B_o, R_s, B_g; B_oi at initial conditions	Differential values used unconverted; B_t and B_o confused below P_b
Gas material balance	Z as a function of pressure at reservoir temperature	Uncorrected Z on a sour gas; two-phase Z from CVD used where single-phase Z belongs
Nodal analysis	μ_o, μ_g, R_s, B_o over the full wellbore pressure and temperature range	Reservoir-temperature viscosity applied up a cooling wellbore
Simulation	Consistent tables, monotonic, extended beyond expected depletion	Tables that stop at the lowest laboratory pressure and are then extrapolated by the simulator, silently
Reserves^[17]	Documented basis, sample provenance, stated uncertainty	A PVT report cited without the sampling conditions that determine whether it means anything

Every property table has an edge. Reservoirs do not stop at it, and simulators do not stop either — they extrapolate, without saying so.

QC checklist before you use a PVT report

Provenance. Which well, which date, which reservoir unit, which sampling method. A report without these is unusable for reserves regardless of its contents.
Flowing bottomhole pressure at sampling, compared against the reported saturation pressure. This is the representativity question, and it is answered here or nowhere.
Producing GOR stability during the sampling period, and for recombined samples, the GOR used and the separator stage it refers to.
Saturation pressure against reservoir pressure. A reported P_b above measured reservoir pressure is not possible for a fluid that was single-phase in place. Resolve it before proceeding.
Contamination. Whether oil-based mud was used, whether contamination was quantified, and whether the tabulated properties are the corrected ones.
Fluid type consistent with GOR, stock-tank gravity and composition — and consistent with the laboratory studies that were actually run.
Y-function plot of the CCE data below P_b. Straight, or explained.
Separator conditions in the test matching the facilities you intend to build.
Conversion to field basis performed, and reproducing the separator test exactly at P_b.
Table range extending below the lowest pressure the reservoir is forecast to reach.
Multiple samples compared where they exist. Two disagreeing reports are information about the reservoir — a compositional gradient, more than one fluid system, or one bad sample — not an inconvenience to be averaged.

None of this is exotic. It is an hour with the report and a calculator, against a number that scales every volume the asset will ever report. The reason PVT errors survive is not that the checks are difficult. It is that the report looks finished when it arrives.

References

McCain, W.D., Jr. The Properties of Petroleum Fluids, 3rd edition. PennWell, Tulsa (2017), ISBN 978-1-59370-373-8. The 2nd edition (PennWell, 1990, ISBN 978-0-87814-335-1) remains the more widely cited and is the source of the standard fluid-type classification figures.
Whitson, C.H. & Brulé, M.R. Phase Behavior. SPE Monograph Volume 20, Henry L. Doherty Series, Society of Petroleum Engineers, Richardson TX (2000). The reference treatment of reservoir fluid phase behaviour, EOS application and PVT data handling.
Danesh, A. PVT and Phase Behaviour of Petroleum Reservoir Fluids. Developments in Petroleum Science 47, Elsevier, Amsterdam (1998). Detailed treatment of laboratory studies and the conversion of differential data to a field basis.
American Petroleum Institute. Sampling Petroleum Reservoir Fluids, API Recommended Practice 44, 2nd edition, April 2003. The recognised standard for sampling procedure, well conditioning and sample handling.
Hoffmann, A.E., Crump, J.S. & Hocott, C.R. (1953). Equilibrium Constants for a Gas-Condensate System. Transactions of the AIME, 198, 1–10. Origin of the log(KP) versus characterisation-factor consistency plot.
Whitson, C.H. & Torp, S.B. (1983). Evaluating Constant-Volume Depletion Data. Journal of Petroleum Technology, 35(3), 610–620, SPE-10067-PA, doi:10.2118/10067-PA. Material-balance treatment of CVD data and its consistency.
Standing, M.B. (1947). A Pressure-Volume-Temperature Correlation for Mixtures of California Oils and Gases. Drilling and Production Practice, API. The original black-oil correlation.
Vazquez, M. & Beggs, H.D. (1980). Correlations for Fluid Physical Property Prediction. Journal of Petroleum Technology, 32(6), 968–970, SPE-6719-PA, doi:10.2118/6719-PA. Note the author's surname appears as both Vazquez and Vasquez across SPE records.
Beggs, H.D. & Robinson, J.R. (1975). Estimating the Viscosity of Crude Oil Systems. Journal of Petroleum Technology, 27(9), 1140–1141, SPE-5434-PA, doi:10.2118/5434-PA.
Glasø, Ø. (1980). Generalized Pressure-Volume-Temperature Correlations. Journal of Petroleum Technology, 32(5), 785–795, SPE-8016-PA, doi:10.2118/8016-PA. Developed on North Sea crudes.
Al-Marhoun, M.A. (1988). PVT Correlations for Middle East Crude Oils. Journal of Petroleum Technology, 40(5), 650–666, SPE-13718-PA.
Standing, M.B. & Katz, D.L. (1942). Density of Natural Gases. Transactions of the AIME, 146(1), 140–149, SPE-942140-G. The original Z-factor chart.
Dranchuk, P.M. & Abou-Kassem, J.H. (1975). Calculation of Z Factors for Natural Gases Using Equations of State. Journal of Canadian Petroleum Technology, 14(3), 34–36, doi:10.2118/75-03-03. The equation-of-state fit to the Standing–Katz chart.
Peng, D.-Y. & Robinson, D.B. (1976). A New Two-Constant Equation of State. Industrial & Engineering Chemistry Fundamentals, 15(1), 59–64, doi:10.1021/i160057a011. Source of the equation and coefficients quoted above.
Pedersen, K.S. & Christensen, P.L. Phase Behavior of Petroleum Reservoir Fluids. CRC Press / Taylor & Francis (2007). Characterisation of the heavy fraction and EOS tuning practice.
Soave, G. (1972). Equilibrium Constants from a Modified Redlich-Kwong Equation of State. Chemical Engineering Science, 27(6), 1197–1203.
SPE, WPC, AAPG, SPEE, SEG, SPWLA & EAGE (2018). Petroleum Resources Management System (PRMS), revised June 2018. Requirement for a documented, auditable technical basis with stated uncertainty.

PVT — from the sample to the number you trust

First decide what fluid you have[1][2]

Sampling: where most PVT error is created[4]

The mechanism of the bias

Well conditioning

Recombination

Oil-based mud contamination

What the laboratory actually does[1][3]

The conversion nobody checks[1][3]

Checking the report for internal consistency[5][6]

The Y-function — smoothness of the CCE data below Pb