100 percent inspection explained: what checking every unit means, when it is required, and why it does not catch everything
100 percent inspection is the examination of every single item in a batch against the acceptance criteria instead of a sample, so that each unit is individually judged conforming or nonconforming before it is released. Written out as hundred percent inspection, and known on the shop floor as screening, sorting or full inspection, it is the fallback when a sampling plan is not defensible.
The name promises more than the method delivers, and that gap is the single most useful thing to understand about it.
When is 100 percent inspection required?
- Critical characteristics. ISO 2859-1 clause 7.5 reserves the right to inspect every item for designated critical classes of nonconformity, and to reject the lot immediately if a single one is found.
- After a lot has been rejected. The same standard, clause 7.6, says a rejected lot may not be resubmitted until every item has been re-examined or retested and the nonconforming ones removed, replaced or corrected. That re-examination is by definition a full screen.
- Containment. A customer complaint or an escape triggers a sort of all suspect stock, at the supplier, in transit and often at the customer's plant.
- New process, new supplier, new tooling. Before there is enough process data to justify a plan, checking everything is the honest position.
- Very small lots and high-value units. When the sampling plan would ask for most of the batch anyway, sampling buys nothing.
Why does 100 percent inspection not find 100 percent of defects?
This is not a modern observation, and it is worth attributing properly. W. Edwards Deming made it the third of his fourteen points: cease dependence on inspection to achieve quality. His argument was that quality cannot be inspected into a product, a line usually traced back to Harold F. Dodge at Bell Labs, and that by the time you are inspecting, the quality is already in the product, good or bad. Building quality in rather than inspecting it in is the whole basis of the shift from screening to process control.
The measured evidence for how fallible human visual inspection is comes from studies you can check. The US Federal Highway Administration's Reliability of Visual Inspection for Highway Bridges (FHWA-RD-01-020, 2001) put 49 practising state bridge inspectors from 25 agencies through the same inspection tasks. Only 68 percent of routine condition ratings fell within one point of the average, and individual assets drew ratings spanning up to five different levels. In manufacturing, Judi E. See's Visual Inspection Reliability for Precision Manufactured Parts (Human Factors, 2015) reported an average hit rate of 85 percent on defective parts alongside a 35 percent false alarm rate on good ones.
What those numbers do to a batch
Take 10,000 units with 2 percent genuinely nonconforming, so 200 bad parts. Screen all 10,000 with a human inspector performing at an 85 percent hit rate and roughly 30 defective units still ship. Apply that study's 35 percent false alarm rate to the 9,800 good units and a large volume of conforming stock gets pulled for rework it does not need. Detection and false alarm rates vary a great deal by task, lighting, defect type and time on shift, so treat the arithmetic as directional rather than as a forecast for your line. The direction is the point: a full screen is a filter with a known leak, not a guarantee.
Why inspecting twice does not fix it
The instinctive response to an escape is a second sort, then a third. If the two passes made independent errors, the escape rate would fall sharply each time. They do not. The defects a first inspector misses are the marginal, low-contrast, awkwardly located ones, and those are exactly the defects the second inspector also struggles with, so the errors correlate. Each additional pass costs full labour, adds more false rejects, and recovers a shrinking share of what was left. A line running triple sort is describing a process problem, not solving one.
What it costs
In the cost of quality model used by ASQ, screening is an appraisal cost: prevention, appraisal, internal failure and external failure are the four buckets, and appraisal is the one that buys information without changing the process that generates the defects. Full inspection on a high-volume line is a permanent labour cost with no expiry date. It also has a way of becoming invisible: once a sort is running, the underlying capability problem stops hurting anyone visibly, so it stops getting fixed.
100 percent inspection compared with sampling
Statistically, sample inspection accepts a known consumer's risk in exchange for a fraction of the effort. A full screen removes the sampling risk and replaces it with inspector error, which is often the larger of the two. Where the check is automated, that trade changes: in-line machine vision, leak testing, weight checking and functional end-of-line test apply the same criterion to every unit without fatigue, and they are where 100 percent inspection comes closest to its name. The limit is that a sensor only finds what it was set up to find, so an automated screen is not a general safety net either.
Either way, the prior question is whether the check can discriminate at all. Measurement system analysis exists for this: gauge repeatability and reproducibility for measured characteristics, attribute agreement analysis for pass or fail judgements. Running a full screen with a gauge or a criterion that cannot reliably separate good from bad multiplies effort without adding information.
Related reading: in process inspection, final inspection, and quality assurance inspection.