AQL is the defect rate a sampling plan is designed to pass most of the time, protecting the producer; LTPD is a worse rate it should rarely pass, protecting the buyer. Both are points on the plan's operating characteristic (OC) curve, which plots chance of acceptance against defect rate.
The quality-level vocabulary comes from MIL-STD-105E (10 May 1989), the public-domain military standard that ANSI/ASQ Z1.4 descends from. Section 3.1 defines the AQL as the quality level that, for sampling purposes, "is the limit of a satisfactory process average." Its note is blunter than most summaries: "The AQL alone does not identify the chances of accepting or rejecting individual lots or batches," and "It is necessary to refer to the operating characteristic curve of the plan to determine the relative risks." Section 4.4.2 closes a common loophole: choosing an AQL "shall not imply that the contractor has the right to supply any defective unit of product."
The buyer's number sits at the other end. The NIST/SEMATECH e-Handbook defines the lot tolerance percent defective (LTPD) as "a designated high defect level that would be unacceptable to the consumer." Fabrico notes it is "also called the rejectable quality level" (RQL), and MIL-STD-105E calls the same idea limiting quality (LQ), always paired with a consumer's risk. Each point carries a risk. The producer's risk, alpha, is the chance a lot at the AQL is rejected. The consumer's risk, beta, is the chance a lot at the LTPD is accepted anyway.
| Point on the curve | Protects | Risk at that point | Values sources cite |
|---|---|---|---|
| AQL (acceptable quality level) | The producer: good lots should pass | Alpha, the chance a lot at the AQL is rejected | Alpha "often set around 5%" (Fabrico) |
| LTPD, also called RQL or limiting quality | The buyer: bad lots should fail | Beta, the chance a lot at the LTPD is accepted | Beta "commonly set around 10%" (Fabrico); MIL-STD-105E prints LQ at 10% and 5% |
NIST gives a wider range for both risks, with typical values from 0.2 to 0.01. The 5% and 10% pair is a convention, not a rule; what matters is that both numbers are written down.
| AQL — Acceptance Quality Limit | LTPD — Lot Tolerance Percent Defective | |
|---|---|---|
| What it is | The worst process average a plan still treats as satisfactory. MIL-STD-105E calls it "the limit of a satisfactory process average," and section 4.4.2 adds that choosing one gives no right to ship defective units. | A high defect level the buyer will not tolerate in a lot. NIST defines it as "a designated high defect level that would be unacceptable to the consumer"; also called the RQL or limiting quality (LQ). |
| Whose risk | The producer's. Producer's risk, alpha, is the chance a lot that sits exactly at the AQL is rejected anyway. | The buyer's. Consumer's risk, beta, is the chance a lot that sits exactly at the LTPD is accepted anyway. |
| Where on the OC curve | Near the top, in the high-acceptance region — by convention about a 95% chance of acceptance (alpha near 5%). | Near the bottom, in the low-acceptance region — by convention about a 10% chance of acceptance (beta near 10%). |
| Plain meaning | The quality level the plan is built to pass most of the time. | The bad quality level the plan is built to catch and fail most of the time. |
Minitab describes the operating characteristic curve as the picture of "the discriminatory power of an acceptance sampling plan" and says "You should always examine the OC curve before using a sampling plan." The horizontal axis is the true percent defective in the lot. The vertical axis is the chance the plan accepts it. A plan is fully described by its sample size n and its acceptance number c, and the curve follows from the binomial formula NIST gives: add up the chances of finding 0, 1, 2 and so on up to c defectives in a random sample of n.
Read the solid curve from left to right. It is the plan MIL-STD-105E Table II-A gives for code letter L at AQL 2.5 under normal inspection: sample 200, accept on 10 or fewer defectives, reject on 11. At the AQL of 2.5% it accepts 98.7% of lots, so its producer's risk is about 1.3%. It accepts half the lots at 5.3% defective, the point MetricGate calls the indifference point, "the defect rate the plan cannot tell apart from either direction." It falls to 10% acceptance at 7.6% defective. That is this plan's LTPD at a 10% consumer's risk, about three times the AQL. Between 2.5% and 7.6% neither side is well protected.
The dashed curve is a different plan, built in the worked example below to hit an AQL of 1% and an LTPD of 5%. It is steeper and sits further left. A lot that is 2.5% defective, comfortably acceptable to the solid plan, passes the dashed plan only 58% of the time. MetricGate sums up the trade-off: "raising n steepens the curve while raising c shifts it right and flattens it."
Section 4.11 of MIL-STD-105E covers the buyer's side directly. For an isolated lot, plans "can be selected by choosing a Limiting Quality (LQ) and a consumer's risk to be associated with it," and "Tables VI and VII give values of LQ for the commonly used consumer's risks of 10 percent and 5 percent respectively." The same section adds that LQ can help when "specifying the AQL and Inspection Levels for a series of lots," which is how a buyer should use it. Here is what those tables say about one AQL as the lot, and therefore the sample, grows. All rows are general inspection level II and normal inspection.
| Lot size (Table I, level II) | Code letter | Sample | Accept on (Table II-A) | Chance of accepting a 2.5% lot (ours) | LQ at 10% acceptance (Table VI-A) | LQ at 5% acceptance (Table VII-A) |
|---|---|---|---|---|---|---|
| 16 to 25 | C | 5 | 0 | 88.1% | 37% | 45% |
| 91 to 150 | F | 20 | 1 | 91.2% | 18% | 22% |
| 281 to 500 | H | 50 | 3 | 96.4% | 13% | 15% |
| 1,201 to 3,200 | K | 125 | 7 | 98.6% | 9.4% | 11% |
| 3,201 to 10,000 | L | 200 | 10 | 98.7% | 7.7% | 8.5% |
| 10,001 to 35,000 | M | 315 | 14 | 98.6% | 6.4% | 7.0% |
Every row is AQL 2.5, yet the buyer's protection differs by a factor of almost six. A 5-piece sample from a lot of 20 still passes a lot that is 37% defective one time in ten. A 315-piece sample from a lot of 20,000 lets through a 6.4%-defective lot only one time in ten. The producer's side moves the other way: the small plan rejects about 12% of lots that sit exactly at the AQL, the large one about 1.4%. This is what section 3.1 means when it says the chance of acceptance at the AQL "depends upon the sample size, being generally higher for large samples than for small ones." Quality Digest's description of Z1.4 plans as giving "(roughly) a 95-percent chance of acceptance" at the AQL is a fair average, not a property of any one plan.
Two cautions. Inspection level and inspection severity are different settings: the level (I, II, III or S-1 to S-4) picks the sample size from the lot size, while normal, tightened and reduced change the accept and reject numbers, and reduced inspection also cuts the sample (200 to 80 for code letter L), and the tables above are for normal inspection only. The AQL sampling note shows how levels I and III move the sample, and the switching rules page covers severity. And these are MIL-STD-105E values; we quote a handful of cells, not the tables, and the edition your contract names governs.
MIL-STD-105E prints an OC curve and a table of points for every code letter (Table X), and section 4.12.1 explains the method: curves for samples of 80 or fewer use the binomial, and curves for larger samples use the Poisson distribution, which the standard calls "an adequate approximation to the binomial distribution under these conditions." The note under Table X-L-1 repeats that all its values use the Poisson approximation. Here is the AQL 2.5 column for code letter L next to our exact binomial figures for the same plan.
| Chance of acceptance | Percent defective, MIL-STD-105E Table X-L-1 | Percent defective, our binomial |
|---|---|---|
| 99% | 2.39 | 2.42 |
| 95% | 3.08 | 3.11 |
| 90% | 3.51 | 3.54 |
| 50% | 5.33 | 5.33 |
| 10% | 7.70 | 7.60 |
| 5% | 8.48 | 8.33 |
| 1% | 10.1 | 9.82 |
The two agree to within a few hundredths of a point near the top of the curve and drift apart in the tail, where the Poisson approximation puts the limiting quality slightly higher. For a buyer the printed value is the more cautious one, so either is fine for a decision. Our Poisson recalculation reproduces every printed cell above, and our binomial recalculation reproduces the Table VI-A and VII-A values for the small samples (80 pieces or fewer), matching the binomial basis section 4.12.1 gives for curves of 80 pieces or fewer. That is a useful check if you ever compute a curve yourself.
Fabrico's article uses a pair of targets without giving the plan behind them: an AQL of 1% with about 95% acceptance and an LTPD of 5% with about 10% acceptance. NIST explains why no table lookup gives this directly: the sample size and acceptance number are the solution of two equations that "are nonlinear so there is no simple, direct solution," and iterative methods are used. The search is short by hand or in a spreadsheet. For each acceptance number, find the smallest sample that accepts a 5%-defective lot no more than 10% of the time, then check the chance of accepting a 1%-defective lot.
| Accept on | Smallest sample with no more than 10% acceptance at 5% defective | Chance of accepting a 1% lot | Meets 95% at 1%? |
|---|---|---|---|
| 0 | 45 | 63.6% | No |
| 1 | 77 | 82.0% | No |
| 2 | 105 | 91.1% | No |
| 3 | 132 | 95.6% | Yes |
So the smallest single sampling plan that meets both points is a sample of 132, accept on 3, reject on 4. By our arithmetic it accepts 95.6% of lots at 1% defective and 9.9% at 5%. The ratio of LTPD to AQL here is 5. MetricGate warns that "a discrimination ratio LTPD / AQL close to 1 demands a very large sample"; ask for an LTPD of 2% with the same AQL and risks and the same search ends at a sample of 1,235, accept on 18. The c=0 sampling plan answer covers the special case where the acceptance number is held at zero, which is the first row of this search.
Minitab's pen example is a useful cross-check. It samples 52 pens from a shipment of 5,000 and reports a 0.957 chance of accepting a 1.5%-defective lot and 0.097 for a 10%-defective lot, without printing the acceptance number. A plan of 52 with accept on 2 reproduces both figures (0.9567 and 0.0966), and running the same search for an AQL of 1.5% and an LTPD of 10% makes 52, accept on 2, the smallest plan that meets both, so the example matches the classic 95% and 10% pair. When Minitab's supervisor cuts the sample to 35, the chance of accepting a 10%-defective lot becomes 0.306. With the same acceptance number, our figure for a 1.5% lot rises to 98.5%. The supplier gains a little and the buyer gives up a lot.
The OC curve tells you which lots pass. The average outgoing quality (AOQ) tells you what the buyer ends up holding over many lots. MIL-STD-105E section 3.2 defines it as the average quality of all accepted lots plus all rejected lots "after the rejected lots or batches have been effectively 100 percent inspected and all defectives replaced by non-defectives," and section 3.3 calls its maximum the average outgoing quality limit (AOQL). NIST gives the formula: AOQ equals the chance of acceptance, times the incoming defect rate, times (N − n) / N, where N is the lot size.
For the code letter L plan on a lot of 5,000, our figures are: 2.5% incoming gives 2.37% outgoing; the worst case, the AOQL, is about 3.1% outgoing when lots arrive near 4% defective; and 10% incoming gives under 0.1% outgoing, because almost every such lot is rejected and sorted. That last figure holds only if rejected lots really are screened piece by piece and defectives replaced. If a rejected lot is simply returned, reworked by eye and resubmitted, the AOQ arithmetic does not apply, and the quality agreement should say which happens.

Plans from ANSI/ASQ Z1.4 and ISO 2859-1 are looked up by AQL and, like MIL-STD-105E (section 4.11), are built for a continuing series of lots from one supplier. Fabrico notes that LTPD-indexed schemes such as the Dodge-Romig LTPD tables suit cases where protecting the consumer from accepting individual bad lots matters most, and MIL-STD-105E section 4.11 points isolated lots to limiting quality. The practical answer for a repeat program is to keep the AQL plan and check it against an LTPD you choose. For a one-off lot, start from the LTPD.
To get the plan to check, start with the AQL calculator: it returns the sample size and accept and reject numbers from your lot size, inspection level and AQL.
Quality Magazine (John Vandenbemden, chair of the ASQ Standards Committee, April 2026) reports the third edition's final draft approved December 2025 and lists among its changes "the removal of operating characteristic (OC) curves for each plan; the standard now includes methods for creating individual OC and Average Sample Number (ASN) curves in Annex E." The same column says editing errors still had to be fixed before publication, so treat any early copy with care. We have not read the third edition. If that change holds, buyers working to it will compute curves the way this page does rather than read them from a chart. MIL-STD-105E's printed curves remain a convenient public check, and the method itself does not change.
Close, but check the definition your contract uses. Fabrico says LTPD is "also called the rejectable quality level", and MIL-STD-105E uses limiting quality (LQ) with a stated consumer's risk of 10 or 5 percent. A reader comment on a 2018 Quality Digest article says the documents AS9138 and ARP9013 now separate them: LTPD at 10% acceptance and RQL at 5%. We have not read those documents, so write the acceptance probability next to the number.
No. MIL-STD-105E section 4.4.2 says the selection or use of an AQL "shall not imply that the contractor has the right to supply any defective unit of product." The AQL is the process average a plan treats as satisfactory, and Fabrico calls it "not a target to aim for."
No. Quality Digest describes Z1.4 plans as giving "(roughly) a 95-percent chance of acceptance" at the AQL, but MIL-STD-105E section 3.1 says the chance "depends upon the sample size, being generally higher for large samples than for small ones." At AQL 2.5 under normal inspection, our binomial figures run from 88.1% for a 5-piece sample to 98.7% for a 200-piece sample.
Look up its code letter and AQL in MIL-STD-105E Table VI-A (10% acceptance) or Table VII-A (5% acceptance), or compute it with the binomial formula on this page. For code letter L at AQL 2.5, the printed values are 7.7% and 8.5%; our binomial figures are 7.6% and 8.3%.
For the buyer, generally yes. At the same AQL the limiting quality falls as the code letter and sample grow: from 37% for a 5-piece sample to 6.4% for a 315-piece sample at AQL 2.5 in Table VI-A. In Minitab's example, cutting a 52-piece sample to 35 raises the chance of accepting a 10%-defective lot from 0.097 to 0.306.
No. It gives the starting point only: the normal-inspection single sampling plan, meaning how many pieces to pull and how many defectives pass or fail the lot. Carry those two numbers into Table VI-A or the binomial formula here to see what the plan does at the defect rate you cannot accept.
Writing a sampling plan into a quality agreement? Share the drawing and the defects that matter to you, and we will help you set the AQL, pick an LTPD you can live with and check that the plan meets both, ahead of the first production lot. Happy to sign an NDA first.
Send a part for reviewBook a 30-minute callHave a question about AQL vs LTPD? Ask our AI assistant. It answers from this website, and our tooling team follows up if you want.
AI assistant. Answers can be wrong; nothing it says is a quote or a promise.