How many practice
questions before
the CPA exam?
Nobody can answer “how many practice questions before the CPA exam” with a number, because the mapping from practice performance to a scaled exam score has never been published by anybody. What can be answered, with the arithmetic printed rather than asserted, is what each additional correct answer buys. In our own estimator, 20 correct answers with no simulation attempted reads near 52% and 200 correct answers with no simulation attempted reads near 76%. The figure moves, and it moves slowly, and below we show exactly why.
beta · not yet validated against real exam outcomes
Source: AICPA, “Learn more about CPA Exam scoring and pass rates” · checked 2026-08-15 · the 42.12% prior and the MCQ / simulation score split
Why the honest answer is not a number
The figures quoted in this market — a target volume of practice work, stated as a range — share one property: none of them names the evidence it rests on. There is no published study relating practice volume to CPA Exam outcomes, no vendor publishes its own predicted-versus-actual data, and the exam itself is scored on a scale whose mapping from performance to reported score the AICPA does not release. A volume target under those conditions is a marketing figure with a plausible shape.
So this page answers a different question, and it is the one that is actually answerable: given a practice record, what does a defensible estimate of passing look like, and how much does the next answered item move it?
How many practice questions before the CPA exam, run through our own arithmetic
The estimator on this site starts every candidate at the national FAR pass rate — 42.12% cumulative in 2025 — and moves away from it in proportion to evidence. Four anchor points fall out of it, and all four are asserted in our test suite so that changing a constant breaks the build rather than quietly changing what the site says:
| Practice record | Estimate |
|---|---|
| Nothing answered at all | 42.1% |
| 20 answered correctly, no simulation attempted | ~52% |
| 200 answered correctly, no simulation attempted | ~76% |
| Both halves answered correctly, simulations included | ~80% |
| Multiple choice correct, simulation cells all wrong | ~19% |
These are outputs of the model described on our methodology page, not observed pass rates. They have not been checked against real exam outcomes.
Read the shape rather than the numbers. The first stretch is steep: going from nothing to 20 correct moves the estimate about ten points, because at that point almost all of the movement is the prior giving way. The next stretch is much flatter: ten times as much correct work adds roughly twenty-four points. And the last row is the one worth sitting with — a candidate who answers every multiple-choice item correctly and every simulation cell wrong reads below the national base rate.
The constants, printed
Three numbers do most of the work, and they are ours rather than anybody else’s:
- The prior is 42.12%, the 2025 cumulative FAR pass rate read off the AICPA’s own pass-rate page. With no evidence about you, that is the estimate.
- The shrinkage strength is 10. Evidence is weighted against a prior worth ten answered items, so the first ten move the figure a long way and the thousandth moves it almost not at all. That is why a volume target is the wrong shape of advice: the return per answer falls away.
- The simulation half is carried at 0.5 when untested, not extrapolated from multiple-choice work, and its own shrinkage strength is 6. Simulations carry about half the FAR score, so leaving that half unmeasured leaves half the estimate resting on an assumption of pure uncertainty.
That last choice is the one that produces the 52% figure people quote back at us, and it is worth being precise about what it does and does not mean. It is not a limit on what a multiple-choice-only record can reach: 200 correct answers with no simulation attempted reads near 76%, which is well above 52%. What it does is slow the climb and keep the interval wide, because for half the exam there is genuinely no data.
What to do with this instead of a target
Count simulation cells, not questions. A five-part simulation is five cells, and cells answered under time are the scarce evidence in almost every candidate’s record. The single largest movement available to most people is not more multiple-choice work; it is the first honest simulation attempt.
Count coverage, not volume. Answering correctly across many Blueprint groups is worth more than the same volume concentrated in three, because an estimate built group by group has to discount the groups it has never seen. The Blueprint hub makes “never tested” countable.
Then stop asking for a target. Put your real record into the estimator and read the interval rather than the point.
Common questions
How many practice questions should I do before the CPA exam?
There is no published figure, and any number quoted without a model behind it is a guess. What can be stated is what additional correct answers buy in an evidence-based estimate. In ours, 20 correct answers with no simulation attempted lands near 52% and 200 correct answers with no simulation attempted lands near 76% — the number rises, and it rises slowly, because half the FAR score sits in simulations that have not been tested.
How many MCQs do I need to pass FAR?
Nobody can tell you a number, because the exam is not scored on a count of practice questions and the mapping from practice performance to a scaled score has never been published by anyone. What the arithmetic below does say is that the return on each additional correct answer falls away quickly, and that answering more multiple-choice questions is the slowest remaining way to move the estimate once the simulation half is still untested.
Does skipping simulations hold the estimate down?
It holds it back rather than holding it down, and the difference matters. Simulations are worth about half the FAR score, and when none has been attempted that half is carried as unknown rather than extrapolated from multiple-choice work. The figure still climbs — 200 correct answers reads near 76% against 52% at 20 — but it climbs more slowly and the interval around it stays wider.
Is this estimate calibrated against real exam outcomes?
No. It has never been checked against predicted-versus-actual exam results, because we do not have that data. The constants below were chosen by hand, the anchors are asserted in our own test suite, and the whole thing is a shrinkage toward a published base rate rather than a fitted model. Read it as arithmetic you can audit, not as a forecast.
Related
How long to study for FAR, from the published weights → · What seven simulations cost you in minutes → · Eight questions to ask any readiness score →
Source: NASBA, “CPA Examination” · checked 2026-08-15 · exam sections and administration