Guide
How do you set an AQL sampling plan for machined parts?
An AQL sampling plan is a rule for accepting or rejecting a whole lot of parts from a random sample: it fixes how many parts to inspect and how many nonconforming parts the sample may contain before the lot is rejected. The plan is looked up from the lot size, an inspection level and an acceptance quality limit (AQL) using the tables in ANSI/ASQ Z1.4 or ISO 2859-1.
By Saola Manufacturing engineering team · Published · Updated
Machined and die-cast parts are not cartons of consumer goods. One drawing can carry a thread that must gauge, a sealing bore, a cosmetic face and a dozen dimensions that barely matter, and each deserves a different level of risk. This guide covers what an AQL promises, how to classify defects, how to read the sample size, when a zero-acceptance plan fits better, and how to write the requirement into a purchase order.
On this page
What does an AQL actually promise?
An AQL is the worst process average that a sampling scheme will still accept most of the time over a continuing series of lots; it is not a count of defective parts the buyer has agreed to receive. It describes the long-run behavior of the plan, not the quality of any single lot.
The definition in MIL-STD-105E, the 1989 edition of the military series from which ANSI/ASQ Z1.4 and ISO 2859-1 descend, makes the point directly: the AQL alone does not identify the chances of accepting or rejecting an individual lot, and the operating characteristic (OC) curve of the plan has to be read to see the actual risks (MIL-STD-105E, paragraph 3.1). The NIST handbook is equally plain about the purpose of lot sampling: it decides whether a lot is likely to be acceptable; it is not a way to estimate the quality of that lot (NIST, What is acceptance sampling?).
So a lot that passes can still contain nonconforming parts, and the AQL is only half the risk picture. The other half is the lot tolerance percent defective (LTPD), the level you would call bad, and how often the plan still accepts a lot that bad; the NIST handbook calls these the producer’s and consumer’s risks (NIST, lot acceptance sampling plans). Setting an AQL without checking the LTPD chooses only the supplier’s side of the bargain.
The schemes also assume a continuing stream of lots with switching rules between normal, tightened and reduced inspection (ASQ, Z1.4 and Z1.9). An isolated order does not get that protection, so read the OC curve or inspect more.
How should critical, major and minor defects be classified on a machined part?
Classify each drawing characteristic by the consequence of getting it wrong: critical if a nonconformance could make the product unsafe, major if it would cause a failure or materially reduce usability, and minor if it has little bearing on fit or function. Each class then gets its own plan.
Those three definitions follow MIL-STD-105E and MIL-STD-1916, which classify defects by seriousness rather than by feature type. Applied to a typical turned or milled part, the classes tend to fall out like this:
- Critical. A thread or seat on a gas or pressure fitting that must seal, a retention feature whose failure lets a part come loose, material or heat treatment on a load-bearing component, and anything a regulator or your end customer names as safety related.
- Major. Mating diameters and bores with a fit tolerance, thread gauging on non-safety threads, datum features, position of holes that locate the part in an assembly, and surface roughness on a sealing or sliding face.
- Minor. Dimensions held to the general tolerance note, chamfers and edge breaks, cosmetic marks on non-visible faces, and part marking legibility.
Put the class on the drawing itself, with a symbol or balloon code next to each critical and major characteristic, so the supplier’s inspector, a third-party inspector and your receiving inspector work from one document. Keep the critical list short; if a third of the drawing is critical, the classification has stopped carrying information.
The same logic applies to a die-cast part machined afterward: porosity breaking into a machined sealing face is major or critical, while as-cast texture on a hidden face is minor. The tolerance guide shows how to separate as-cast and machined tolerances on the drawing.
How do you find the sample size for a lot of machined parts?
Look up the lot size against the inspection level in the sample size code letter table, then read the sample size and the accept and reject numbers for your AQL from the single sampling table for normal inspection. General inspection level II is the default unless you specify another.
The table below reproduces the general inspection level II column of the code letter table and the sample size for each letter. Levels I and III give smaller and larger samples for the same lot size, and the four special levels, S-1 to S-4, are meant for costly or destructive tests where a small sample is unavoidable.
| Lot or batch size | Code letter (level II) | Sample size |
|---|---|---|
| 2 to 8 | A | 2 |
| 9 to 15 | B | 3 |
| 16 to 25 | C | 5 |
| 26 to 50 | D | 8 |
| 51 to 90 | E | 13 |
| 91 to 150 | F | 20 |
| 151 to 280 | G | 32 |
| 281 to 500 | H | 50 |
| 501 to 1,200 | J | 80 |
| 1,201 to 3,200 | K | 125 |
| 3,201 to 10,000 | L | 200 |
| 10,001 to 35,000 | M | 315 |
| 35,001 to 150,000 | N | 500 |
| 150,001 to 500,000 | P | 800 |
| 500,001 and over | Q | 1,250 |
The sample does not grow in proportion to the lot, so sampling gets relatively cheaper as orders grow. Where the accept and reject columns show an arrow, use the first plan above or below it, sample size included; if that sample equals or exceeds the lot, the standard calls for 100 percent inspection.
Define the lot, too: parts from the same material, process route and conditions. A shipment mixing several runs or bar heats is a box, not a lot, and sampling it as one hides the run that went wrong.
What does a worked AQL example look like for turned parts?
For a lot of 1,000 turned fittings at general inspection level II, the code letter is J and the sample is 80 parts, inspected against a separate accept number for each defect class. With majors at AQL 1.0 and minors at AQL 2.5, the lot passes on no more than two major and five minor nonconforming parts.
Here is the full plan, read from MIL-STD-105E Table II-A for normal inspection, for a CNC-turned brass fitting with a sealing thread:
- Critical: the sealing thread and the sealing face. Not sampled with an AQL at all. Every part is gauged, or the supplier runs a fail-safe or automated check, and one nonconformance stops the lot for containment.
- Major, AQL 1.0. Sample 80, accept on 2, reject on 3. Covers the fit diameters, the locating shoulder and thread gauging on the non-sealing end.
- Minor, AQL 2.5. Sample 80, accept on 5, reject on 6. Covers general-tolerance dimensions, chamfers and cosmetic condition.
The same 80 parts can be used for the major and minor checks. Each class is judged on its own count, and a failure in either class rejects the lot. A reject does not mean scrap: typical industry practice is that the supplier screens the rejected lot 100 percent, removes the nonconforming parts, and resubmits it, which is also the assumption behind the average outgoing quality figures in the standards.
Before signing off, check what the major plan really does; the next section shows it still accepts most lots that are twice as bad as the AQL.
When should you use a c=0 plan instead of a standard AQL plan?
Use an accept-on-zero (c=0) plan when any nonconforming part found in the sample should reject the lot, which fits major characteristics on functional machined features and the verification sample for screened critical characteristics. The cost is a higher chance that a good lot gets rejected.
MIL-STD-1916, the Department of Defense standard for preferred acceptance methods issued in 1996, accepts a lot only if no nonconforming units are found in the sample, and its foreword states that it follows the DoD policy of eliminating AQLs from specifications. For critical characteristics it goes further, requiring automated screening or a fail-safe operation, with sampling used only to verify that the screening works. The table compares the acceptance behavior of three plans that all use a sample of 80.
| Plan (n = 80) | 0.5% in lot | 1% in lot | 2% in lot | 5% in lot |
|---|---|---|---|---|
| Accept on 0 (c=0) | 67% | 45% | 20% | 2% |
| Accept on 1 (AQL 0.65) | 94% | 81% | 52% | 9% |
| Accept on 2 (AQL 1.0) | 99% | 95% | 78% | 23% |
Read the bottom row first. Under the AQL 1.0 plan, a lot in which two parts in every hundred are nonconforming is still accepted about four times out of five, and even a lot with five in every hundred gets through almost a quarter of the time. The zero-acceptance plan cuts those odds sharply, but it also rejects more than half of the lots that sit exactly at one in a hundred. That trade is why c=0 plans work best with a capable process: a supplier whose process runs well below the reject line rarely trips a zero-acceptance plan, while a marginal process trips it constantly.
Ready-made zero-acceptance tables exist, including the plans in MIL-STD-1916 and the zero-acceptance-number tables that ASQ publishes in book form, so a c=0 plan does not have to be derived from scratch. Whichever you use, name the table and its edition in the purchase order so the supplier can reproduce the sample size.
How should an AQL requirement be written into a purchase order?
Name the standard and its edition, the inspection level, the plan type, and an AQL or accept number for each defect class, and say who inspects, where, and what record comes back with the shipment. Anything left out will be filled in by the supplier’s default.
Government specifications show how short a sound clause can be. MIL-DTL-5541F, for example, requires sampling to ASQ Z1.4 at inspection level II unless the contract says otherwise, and caps a lot at one day’s production of like parts. A commercial purchase order for machined parts needs a little more, because it has to carry the defect classes. A workable template reads:
“Lot acceptance per ANSI/ASQ Z1.4-2003 (R2018), single sampling, normal inspection, general inspection level II, with switching rules applied across consecutive lots. Characteristics marked ‘C’ on drawing rev. D are critical: 100 percent inspection, zero nonconformances accepted. Characteristics marked ‘M’ are major: AQL 1.0. All other characteristics are minor: AQL 2.5. A lot is one production run from one material heat. The supplier ships an inspection record per lot giving the lot size, sample size, accept and reject numbers, and the count of nonconformances per class.”
Adjust the edition to the one your quality system actually holds. ISO published a third edition of ISO 2859-1 in 2026 that replaces the 1999 edition and its amendment and adds skip-lot procedures, so a European buyer may prefer to cite it (ISO 2859-1:2026). Citing both standards in one clause invites an argument about which governs; pick one.
Put the same clause and classification in the RFQ; the RFQ package guide covers what else to send.
What changes when machined parts are sourced from an overseas supplier?
The sampling plan has to work at two points, at the supplier before shipment and at your dock after it, so both sides should inspect to the same written plan and exchange the records. A rejection found after an ocean crossing costs far more to resolve than one found at the factory.
For parts made in Vietnam, Mexico or anywhere far from the assembly line:
- Move the decision upstream. Ask for the supplier’s lot inspection record before the goods are released for shipment, or use a third-party inspector at origin working from the same drawing, classification and table.
- Make receiving inspection a check, not a second verdict. Once the supplier’s records have held up for a run of lots, move receiving to a reduced plan, as the switching rules intend.
- Agree on the gauges. Many disputes over sampled lots are gauge disputes. Name thread gauges by standard and class, and say how each critical and major feature is measured, so an accept at the factory means an accept at your dock.
Saola’s own ISO 9001 certification is in progress and not yet held, so a buyer should specify the sampling plan and records it needs in the order rather than rely on a certificate. The quality and inspection page describes the current setup. To have a part quoted against your sampling clause, send the drawing and the clause with the RFQ, and any characteristic that needs a different plan is raised in the reply.
Frequently asked
Is AQL 1.0 the same as allowing one defective part in a hundred?
No. It means the plan will accept, most of the time, lots from a process running at about that level over a continuing series. Any single lot can be better or worse, and the plan’s OC curve shows how often a worse lot still passes.
Are ANSI/ASQ Z1.4 and ISO 2859-1 interchangeable?
They share the same origin in the MIL-STD-105 series and the same basic structure of code letters, AQLs and switching rules. They are still separate documents with their own editions, so cite one of them in the order and let that one govern.
Should dimensions be sampled by attributes or by variables?
Attribute plans (Z1.4, ISO 2859-1) only count parts as good or bad. Variables plans such as ANSI/ASQ Z1.9 use the measured values and need smaller samples, but they assume the measurements are normally distributed, so they suit stable, measured dimensions rather than go/no-go checks.
What happens to a lot that fails sampling?
The purchase order should say. The usual commercial arrangement is that the supplier screens the whole lot, removes nonconforming parts, corrects the cause and resubmits, with the next lots moving to tightened inspection under the switching rules.
Sources
- ISO 2859-1:2026, Sampling procedures for inspection by attributes — Part 1: Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot inspection, ISO, 2026 (third edition)
- ISO 2859-1:1999, second edition, ISO, 1999 (withdrawn, replaced by the 2026 edition)
- ANSI/ASQ Z1.4-2003 (R2018) and Z1.9-2003 (R2018) overview, ASQ, accessed September 2026
- MIL-STD-105E, Sampling Procedures and Tables for Inspection by Attributes, U.S. Department of Defense, 10 May 1989 (Table I, Table II-A, paragraph 3)
- MIL-STD-1916, DoD Preferred Methods for Acceptance of Product, U.S. Department of Defense, 1 April 1996 (reactivated by Notice 2, 2014)
- MIL-DTL-5541F, Chemical Conversion Coatings on Aluminum and Aluminum Alloys, U.S. Department of Defense, 2006 (paragraph 4.3.1)
- NIST/SEMATECH e-Handbook of Statistical Methods, section 6.2, Lot acceptance sampling, NIST, 2012, accessed September 2026