VE Table Tuning: A Step-by-Step Correction Workflow That Converges

The volumetric efficiency table is the ECU's model of how well your engine breathes. Get it right and fueling falls into place across the whole map; get it wrong and you spend months chasing symptoms in tables that were never the problem.

As of July 2026, VE correction is also one of the most commonly mishandled jobs in DIY tuning — not because the concept is difficult, but because the workflow is where the difficulty lives. Correct cells in the wrong order, or with too little data behind them, and each pass undoes some of the last one.

Here is a loop that converges.

What the table represents

Volumetric efficiency is the ratio of the air mass actually drawn into a cylinder to the mass that would fill it at ambient conditions. A well-breathing engine at its torque peak might exceed one hundred percent through intake tuning; a throttled engine at light load sits far below.

In a speed-density calibration, the ECU has no direct airflow measurement. It reads manifold pressure and air temperature, looks up VE for the current speed and pressure, and computes airflow from those. Fuel follows directly from airflow.

Which produces the property that makes this table foundational: a VE error becomes a fueling error of the same proportion, immediately, everywhere in that region. Eight percent low on VE is eight percent lean, unless closed-loop correction is masking it — and at full load nothing is masking it.

Whether your calibration works this way at all depends on the air model; the MAF versus speed-density comparison covers how to determine which one is deciding on your vehicle, and many calibrations blend both.

Before you touch a cell

Every one of these prerequisites exists because skipping it makes VE data describe something other than volumetric efficiency.

The engine must be mechanically sound. VE is a breathing measurement. A vacuum leak, a stuck PCV valve, a burnt valve or an exhaust restriction all change breathing, and correcting cells around them bakes the fault into the calibration. The classic leak signature — large positive fuel trims at idle that shrink as load rises — should be resolved with a wrench, not a table.

Injector characterization must be right. The ECU converts airflow into a pulse width using its model of injector flow. If injector data is wrong, the pulse width is wrong even from a perfect airflow estimate — and the symptom looks exactly like a VE error. Correcting VE to compensate for bad injector data produces a table that is wrong in a way that only appears when something else changes. The injector data guide covers slopes, offset and breakpoint.

Sensors must be trustworthy. Manifold pressure, intake air temperature, coolant temperature and your wideband. Every VE correction is derived from those readings, so a lazy sensor produces a table that describes the sensor.

You need a wideband for the load cells. Part-throttle closed-loop cells can be corrected from trims, because the ECU reports its own error. High-load cells run open loop with no feedback whatsoever. Without a wideband you can only correct the half of the table that matters least.

The workflow

Step 1 — Log with the table axes in mind

Log engine speed, manifold pressure, both fuel trims, commanded and measured air-fuel ratio, intake air temperature and coolant temperature. The goal is that every sample can be placed in the cell it belongs to, which means the log must be dense enough to populate cells rather than skate across them.

Drive to fill cells deliberately: steady cruise at several speeds and loads, gentle part-throttle acceleration, and separately a full-load pull in one gear for the high-load region. Warm engine throughout — a cold engine has different breathing and enrichment behaviour, and mixing warm and cold data poisons the whole exercise. The channel selection guide covers the trade-off between how many channels you log and how fast you sample them.

Step 2 — Compute error per cell, then discard the thin cells

For each cell, gather every sample that landed in it and compute the average error — from trims where the ECU was in closed loop, from measured versus commanded AFR where it was not.

Then apply the rule that separates a converging workflow from a thrashing one: discard cells with too few samples. A cell with three samples and a wandering error is noise. Correcting it introduces a distortion you will chase for weeks. Correct only cells you actually populated with enough data to trust, and leave the rest for a later log that visits them properly.

Step 3 — Apply proportional corrections

The correction is proportional, not a fixed step. If a region needed roughly eight percent more fuel to reach its target, the airflow estimate there was about eight percent low, so those cells increase by about that proportion.

The MAF and VE correction calculator does this arithmetic from a trim or AFR error and returns the adjusted values, which removes the most common source of transcription mistakes.

Two disciplines matter here:

Correct in regions, not scattered cells. Airflow is continuous — real errors appear across neighbourhoods of cells, not in isolated ones. If a single cell disagrees strongly with all its neighbours, suspect the data before the engine.

Do not chase the last percent on pass one. Get within a few percent, log again, refine. Trying to land exactly on target in a single pass amplifies noise into the table.

Step 4 — Smooth with judgement

A real VE surface is smooth because the physics is smooth. Isolated spikes are usually artifacts.

But smoothing is a blunt instrument. Around a camshaft phasing transition, a resonance point, or the boundary where a variable intake changes state, the engine genuinely changes behaviour and the table should have structure there. Smoothing that flat destroys real information.

Smooth to remove noise. Never smooth to remove shape.

Step 5 — Re-log and repeat

New log, same conditions, same analysis. Errors should be markedly smaller. Two or three passes over the cells you actually drive in is typical.

If errors are not shrinking, stop editing. A table that will not converge is telling you the error is not in the table — go back to leaks, injector data, fuel pressure and sensor health.

Symptom during VE tuning	Likely real cause	Action
Corrections shrink each pass	Working as intended	Continue to convergence
Same cells need correcting every pass	Data conditions differ between logs	Standardize temperature and driving
Whole table needs a uniform shift	Injector scaling or fuel pressure	Fix injector data, not VE
Idle region will not settle	Vacuum leak or PCV	Mechanical diagnosis
Low load fine, high load lean	Fuel delivery limit or duty cycle	Check injector duty cycle
Table looks jagged after edits	Thin cells were corrected	Discard low-sample cells, re-log

The third row deserves emphasis: a uniform error across the entire table is almost never a VE problem. VE errors vary by region because breathing varies by region. A flat offset everywhere points at something global — injector slope, fuel pressure, or a units mistake.

The errors that cost the most time

Correcting cells you barely visited. The single biggest source of tables that never settle.

Mixing hot and cold data. A warm-up log and a fully-warm log describe different engines. Pick one state and stay in it.

Editing VE to fix an injector problem. Produces a table that is wrong in a hidden way, and the hiding stops the moment injectors, fuel pressure or fuel type change.

Ignoring intake air temperature drift. Air density changes with temperature. Comparing a cool first pull to a heat-soaked fourth pull attributes an air effect to a table error.

Changing something else in the same flash. VE correction is a measurement exercise. Any other change in the same file makes the next log ambiguous.

When it is finished

The table is done when a fresh log across your normal operating range shows small, unbiased errors — no consistent lean or rich trend in any region — and full-load pulls show delivered air-fuel ratio tracking commanded within a couple of tenths. At that point the ECU's model of your engine matches your engine, and every downstream decision, including timing, rests on something real.

That is the whole reason this table comes first. The complete guide to tuning your car places it in the wider sequence, and the datalog analysis page covers how TuneVault reviews a log for per-region fueling error, knock retard and trims and returns the specific cells to change with the values to put in them.

If you would rather have your VE corrections checked before the next flash than diagnosed after it, start here.

Frequently asked questions

What does the VE table actually do?

It is the ECU's model of how efficiently your engine fills its cylinders at each combination of engine speed and manifold pressure. From that estimate the controller calculates airflow, and from airflow it calculates how much fuel to inject. If the VE table is wrong in a region, every fuel decision in that region is wrong by the same proportion — which is why VE correction is the foundation of speed-density calibration rather than a refinement of it.

How do I know which VE cells to correct?

By logging error against the cells themselves. You want engine speed, manifold pressure, fuel trims at part throttle and measured air-fuel ratio at load, arranged so each sample lands in the cell it belongs to. Cells with many samples and a consistent error are correctable; cells with two samples and a noisy error are not yet, and editing those is how people introduce problems they later cannot find.

How much should I change a VE cell by?

By the proportion of the measured error, not by a fixed step. If the ECU needed roughly eight percent more fuel to hit its target in a region, the airflow estimate in that region was about eight percent low, so the cells move by about that proportion. Applying a fixed number of counts to every cell ignores that the error is proportional, and it distorts the shape of the table.

Should I smooth the VE table after editing?

Lightly, and with judgement. A real VE surface is smooth because airflow physics is smooth, so a single wildly different cell surrounded by consistent neighbours is usually a data artifact rather than a real feature. But aggressive smoothing across a region where the engine genuinely changes behaviour — around cam transitions, for instance — erases real structure. Smooth to remove noise, never to remove shape.

How many iterations does VE tuning take?

Usually two or three passes over the cells you actually drive in, with a full log between each. Corrections shrink each pass as the model converges. If errors are not shrinking after a few passes, stop correcting cells and question the inputs — a leak, a failing sensor or an injector data error will produce a table that never settles because the error is not in the table.

Do I need a wideband to tune a VE table?

For the load regions, yes. Part-throttle closed-loop cells can be corrected from fuel trims because the ECU is telling you its own error. High-load cells run open loop with no feedback at all, so the only way to know what was delivered is a wideband sensor. A VE table corrected only from trims will be accurate where you cruise and unverified where it matters most.

What if my VE table looks correct but fueling is still wrong?

Then the error is probably not in the VE table. Injector characterization is the usual culprit — if the ECU's model of injector flow is wrong, it will calculate the wrong pulse width from a perfectly accurate airflow estimate. Fuel pressure differing from the assumption in the calibration produces the same symptom. Both look like a VE problem and neither is fixed by editing VE cells.