AQL Sampling for Injection Molded Parts: Lot Size and Acceptance Decisions

AQL sampling inspects a random sample from a molded lot, counts nonconforming parts, and compares that count with the acceptance and rejection numbers set by the chosen plan. Lot size, inspection level, and AQL select the plan. The decision applies to that lot under that plan, and it does not certify every part as good.

What Is Acceptance Sampling by Attributes?

Attribute inspection sorts each sampled part into one of two states against a written requirement: conforming or nonconforming. A short shot on a housing is an attribute result. A wall thickness of 2.41 mm is a variable measurement, though it becomes an attribute result once it is judged only as in or out of tolerance.

Lot acceptance adds a decision rule. The inspector does not estimate a defect rate and then decide by feel. The plan states in advance how many nonconforming parts the sample may contain before the lot is accepted, and that is what separates sampling from a quick look at a few parts from the top of a box.

The National Institute of Standards and Technology describes the purpose in plain terms in its lot acceptance sampling handbook section: deciding lot by lot whether to accept a lot as likely to meet requirements or reject it as likely to have too many defective units. The word “likely” carries weight. A sample can mislead in either direction.

What AQL Means

AQL stands for acceptance quality limit in ISO 2859-1:2026, the third edition, which replaced the 1999 edition and its amendments. Older documents and NIST’s handbook use “acceptable quality level.” NIST’s definition of plan terms treats it as a percent defective that serves as the producer’s baseline, with the plan designed to give a high probability of accepting a lot at or below that level.

Two misreadings follow from the name. AQL 1.0 does not mean each lot may contain 1 percent defects, and it does not mean the buyer has agreed that 1 percent of delivered parts may be bad. It is an index into a sampling scheme, built around long-run process quality. A single lot at exactly that percent defective can still be rejected, which the same NIST page calls producer’s risk.

U.S. contracts may cite ANSI/ASQ Z1.4 rather than ISO 2859-1. Read the standard and edition named on the purchase order, because that document controls the tables you use.

Defining the Inspection Lot

A lot should represent parts made under the same conditions, since the sample speaks only for the population it was drawn from. For molded parts, the boundaries worth stating are the mold, the resin lot, the machine, the process settings, and the production window.

Traceability makes those boundaries usable. If a lot label cannot be traced to a resin lot and a date range, an inspector has no way to isolate affected parts after a rejection.

Mixed lots weaken the result. Suppose a lot combines output from before and after a mold repair. A clean sample could come mostly from the repaired period while the suspect parts sit unsampled. On a multi-cavity tool, a lot also blends cavities, so cavity-specific defects need a sampling approach that reaches every cavity. Splitting such a lot before sampling is a decision to make in writing, not at the receiving dock.

Selecting the Sampling Plan

Five choices define the plan:

1. The applicable standard and edition, or a plan written directly into the contract.
2. The inspection level, which sets the relationship between lot size and sample size.
3. The lot size.
4. The AQL value, which can differ by defect class. A cracked boss and a light flow line do not have to share one value.
5. The plan type: single, double, or multiple sampling.

ISO describes its 2026 edition as defining AQL-indexed single, double, or multiple sampling schemes and as working best alongside ISO 2859-0 for illustrative examples. It also adds skip-lot procedures, according to ISO’s abstract. Anyone working from an older internal procedure should check which of those features the procedure actually assumes.

Sample Size and Acceptance/Rejection Numbers

Ac is the largest count of nonconforming units that still accepts the lot. Re is the smallest count that rejects it. In a single-sampling plan, Re sits one above Ac.

These numbers come from the licensed standard or from the contract. They are not reproduced here, and values copied from a web chart should be checked against the document your contract cites.

Sampling is done without replacement, and a correct sample size does not rescue a biased draw. Pulling 80 parts from the first open carton gives a sample of the carton, not of the lot. A workable method assigns numbers to cartons and positions, then uses a random number source to choose which to open.

Normal, Tightened, and Reduced Inspection

Attribute schemes of this family include normal, tightened, and reduced inspection, with switching rules that move a supplier between them based on recent lot results. Tightened inspection raises the burden after quality slips. Reduced inspection lowers it after a sustained record.

The rules are part of the scheme, not an optional extra. A vignette for the AQLSchemes R package, by the package’s author, describes ANSI/ASQ Z1.4 as derived from MIL-STD-105E and says the switching rules “must be used if the standard is to be properly applied.” A shop that reads one normal-inspection plan off a chart and applies it forever is not running the scheme that the tables were built for.

Who may authorize a switch is a separate question from what the rule says. The agreement should name that authority, because a switching rule applied informally by a receiving clerk leaves no defensible record if a later lot fails in the field.

The switching thresholds themselves belong to the standard you hold. Using them without the current text is guesswork.

When Zero-Acceptance Plans Apply

A c=0 plan accepts a lot only if the sample contains no nonconforming units. One defect found rejects the lot or triggers whatever disposition the agreement requires.

This is a customer-specific or plan-specific requirement, not a general rule for automotive or any other buyer. Check the customer’s quality manual or purchase order. A zero count in a sample also does not show that the lot is free of defects. With a hypothetical sample of 80 and an acceptance number of zero, a large lot that is 1 percent nonconforming still passes about 45 percent of the time (0.99 to the 80th power).

Worked Inspection Decision

The numbers below are hypothetical and were not read from any standard’s table. Assume a contract states its own plan directly: lot of 10,000 molded lids, sample of 80, Ac = 2, Re = 3.

Inspectors draw 80 lids at random and find 2 with short-shot corners and none with other defects. The count of 2 equals Ac, so the lot is accepted. The 2 nonconforming lids are still handled as nonconforming parts, and acceptance does not make them good.

Had they found 3, the count would reach Re and the lot would be rejected, with disposition set by the agreement: return, sort, or concession.

A binomial calculation shows what this plan actually does, treating the lot as large relative to the sample:

True percent nonconforming	Probability the lot is accepted (n = 80, c = 2)
0.5 percent	99.2 percent
1.0 percent	95.3 percent
2.5 percent	67.7 percent
5.0 percent	23.1 percent

Read the bottom row twice. A lot with 5 percent nonconforming parts, or about 500 bad lids, passes this plan in nearly one attempt out of four. That is the arithmetic of the plan, not a defect in the inspector’s work, and it is why the plan choice deserves engineering attention.

The count is one step of a longer chain: define the lot, name the plan, draw a random sample, inspect to written definitions, compare the count with Ac and Re, and record the result. A wrong lot definition or a vague defect standard damages the decision before anyone reaches the count.

What AQL Does Not Prove

An accepted lot does not show that every part is free of defects. As the table shows, an accepted lot can contain a meaningful fraction of nonconforming parts.

It does not show that all dimensional requirements are met, unless those characteristics were in the inspection and were measured on the sample. A visual AQL says nothing about a bore diameter nobody checked.

It does not show that the process is capable. Capability is a statistical statement about process variation over time, drawn from measured data. It does not show PPAP approval either, which is a separate customer decision about the production process and its records, covered in how PPAP submission levels differ for automotive plastic parts.

AQL Sampling FAQ

Can a rejected sample cause a lot to fail even if most parts appear acceptable?

Yes. In the worked plan, 3 nonconforming lids out of 80 reject the lot while 77 sampled lids, or about 96 percent, conform. The plan judges the count against a preset limit. The result does not say that the other 9,920 lids resemble the bad ones, so a rejected lot may be sorted if the agreement allows it.

Is AQL sampling the same as process capability analysis?

No. Sampling judges a finished lot by counting attribute results. Capability analysis uses measured variable data from the process to compare its spread and centering with the tolerance. A process can pass lots while running close to a limit, and that drift would appear in capability data before an attribute count catches it.

Writing a Plan You Can Defend at the Receiving Dock

Start by naming the standard, edition, inspection level, and AQL for each defect class, then write the lot definition next to them so the plan and the population stay tied together. Run the operating numbers for the sample size and acceptance number you intend to use, as the table above does. If a 5 percent nonconforming lot passing at a rate near one in four is unacceptable for this part, the plan is wrong for the part, and a larger sample or lower acceptance number has to replace it. Treat an accepted lot as a decision about that lot under that plan, and keep process capability and customer approval as separate evidence.