Sample inspection
What is sample inspection: sample inspection explained, from lot size to accept and reject numbers
Sample inspection is the practice of checking a defined random subset of a lot and using that result to accept or reject the whole lot, on the basis of a published sampling plan that fixes both the sample size and the number of nonconforming items allowed. The plan sets the pass mark before anyone looks at a part. The inspector counts.
You will also see it called sampling inspection, batch sampling, or AQL inspection. The formal name for the discipline is acceptance sampling by attributes, where every item examined is simply classed conforming or nonconforming rather than measured. Where the characteristic is measured and the decision uses the mean and spread, that is sampling by variables, covered by ISO 3951 instead.
How does sample inspection work?
- Define the lot. A lot is a quantity produced or delivered under conditions presumed uniform. This step gets skipped more than any other, and it matters: pooling two production runs or two suppliers into one lot removes the basis for reading the sample as representative.
- Choose the inspection level. ISO 2859-1 uses level II unless something else is specified. Level I gives less discrimination for less cost, level III more discrimination for more.
- Look up the sample size code letter. Lot size plus inspection level gives a letter.
- Look up the plan. Code letter plus the agreed AQL gives the sample size, the acceptance number (Ac) and the rejection number (Re).
- Inspect and count. Nonconforming items equal to or fewer than Ac, the lot is accepted. Equal to or greater than Re, the lot is not accepted.
- Disposition the lot that failed. ISO 2859-1 clause 7.2 leaves this to the responsible authority: scrapped, sorted, reworked, re-evaluated against more specific usability criteria, or held pending information.
Which standard publishes the sampling plans?
The tables people actually use are published, not invented in house. ISO 2859-1 is the international standard for sampling schemes indexed by acceptance quality limit; the 1999 edition was the working reference for a quarter of a century and was revised in 2026. ANSI/ASQ Z1.4 is the North American equivalent, descended from MIL-STD-105E, and the common plans line up closely with the ISO tables. The rest of the family covers the awkward cases: ISO 2859-2 for isolated lots indexed by limiting quality, ISO 2859-3 for skip-lot procedures, and ISO 2859-5 for sequential plans.
Sample inspection: a worked example
A lot of 2,000 pressed steel brackets arrives. The contract specifies general inspection level II, normal inspection, AQL 2.5. A lot size of 1,201 to 3,200 gives sample size code letter K, and code letter K gives a sample size of 125. At AQL 2.5 the single sampling plan for that letter has an acceptance number of 7 and a rejection number of 8. So: draw 125 brackets at random, inspect them, and count. Six nonconforming, the lot is accepted. Nine nonconforming, the lot is not accepted, and all 2,000 go to disposition, not just the nine.
What AQL means, and what it does not
ISO 2859-1 defines the acceptance quality limit as the "quality level that is the worst tolerable process average when a continuing series of lots is submitted for acceptance sampling". Two things follow from that wording. AQL is a property of a process across many lots, not an allowance you spend on one delivery. And the standard adds the caution itself: designating an AQL "does not suggest that this is a desirable quality level".
The honest framing is narrower than most quality plans imply. A sampling plan controls the risk of accepting a bad lot. It does not certify that an accepted lot is good. Every plan has an operating characteristic curve, plotting probability of acceptance against true lot quality, and that curve always has tails: some genuinely poor lots get through (consumer's risk) and some acceptable lots get rejected (producer's risk). Accepting 2,000 brackets on a sample of 125 means the evidence did not justify rejection. It says nothing about the 1,875 nobody looked at.
The switching rules most people skip
ISO 2859-1 is a scheme, not a single table lookup, and the switching rules are where its pressure on the supplier actually comes from. Two out of five consecutive lots not accepted on original inspection, and inspection switches to tightened. Five consecutive lots accepted under tightened inspection, and it returns to normal. A switching score of at least 30, with steady production, allows a move to reduced inspection. If five lots are not accepted while on tightened inspection, the standard requires inspection to be discontinued altogether until the supplier improves the process and the responsible authority agrees the fix is likely to work.
Pulling one plan out of the tables and running it unchanged forever is common practice. The standard is blunt about it: used that way, without the switching rules, it "shall not be referred to as inspection in compliance with ISO 2859-1". It is then just a repository of individual plans, and the risks have to be assessed plan by plan.
When sampling is the wrong tool
- Critical characteristics. Clause 7.5 reserves the right to inspect every item for designated critical classes of nonconformity and to reject the lot immediately if one is found. Safety-related features usually belong here, which pushes you to 100 percent inspection.
- Very small lots. When the plan asks for a sample that is most of the batch, sampling saves nothing.
- Non-homogeneous lots. Mixed batches break the assumption the tables rest on.
- Destructive or expensive tests. Sequential or variables plans usually reach a decision on fewer items.
- Regulated product needing a record per unit. No sampling plan produces that.
Related reading: quality assurance inspection, incoming goods inspection, and inspection frequency.
