1How to qualify visual inspectors: the protocol.
A manual inspector qualifies by inspecting a controlled test set under production conditions and detecting its defects to criteria approved in advance. FDA’s draft guidance and EU GMP Annex 1 say what the protocol should contain. Neither sets a pass mark.¹²
| Element | What to set | Source |
|---|---|---|
| Eligibility | Normal near visual acuity, with usual lenses; no colour vision impairment; eyesight checks | FDA draft V.C;¹ Annex 1 8.31² |
| Training | The relevant CGMP, the procedure and the defect library; retraining or requalification as appropriate | FDA draft V.C, citing 211.25;¹ Annex 1 8.30² |
| Test set | Good and defective units; defective units typically no more than 10–20%; identities masked | FDA draft V.C¹ |
| Defects in the set | Drawn from a library of all known defect classes, including units at the lower detection limit | FDA draft V.C;¹ Annex 1 8.30² |
| Control of the set | Prepared and approved by QA; the quality unit guards it against manipulation | FDA draft V.C¹ |
| Conditions | 2,000 to 3,750 lux; black and white backgrounds; worst-case time, line speed, container size and fatigue | USP <790> and Ph. Eur. 2.9.20;³ Annex 1 8.31;² FDA draft V.A.3¹ |
| Pace | The qualified rate used in production; compendial reference 5 seconds against each background | Annex 1 8.31;² USP <790>³ |
| Breaks | Distractions minimised; frequent breaks; the longest inspection period without a rest break, in the procedure | Annex 1 8.31;² FDA draft V.A.3¹ |
| Grading | How results are graded and recorded; acceptance criteria | FDA draft V.C¹ |
| Failure | What happens to an inspector who fails | FDA draft V.C¹ |
| Requalification | Method and frequency; at least annually in the EU | FDA draft V.C;¹ Annex 1 8.31² |
Who does what: the quality unit approves the protocol, its acceptance criteria and the requalification frequency, and controls the test sets, which QA staff prepare and approve.¹ A pool of reference inspectors sets the baseline once, by inspecting the test set many times; that baseline becomes the site’s standard for qualifying inspectors.⁴
The light and backgrounds in the table are compendial. USP <790>, Ph. Eur. 2.9.20 and the Japanese Pharmacopoeia’s 6.06 share the light (JP asks 8,000 to 10,000 lux for plastic containers) and backgrounds and allow 10 seconds per container, according to a 2022 presentation by consultant John G. Shabushnig hosted on FDA’s website.³ Annex 1 asks only for “suitable and controlled conditions of illumination and background”, and FDA’s draft asks manufacturers to consider container colour, size and shape when setting the intensity.²¹ In Europe the visible particles test has been a release test for liquid parenteral preparations since the Parenteral preparations monograph (0520) took effect on 1 July 2021, with general chapter 5.17.2 as non-mandatory guidance on inspection.⁵
1.1Test kit and qualification run: typical figures.
No regulator sets these numbers. They are the practice Shabushnig’s FDA-hosted slides describe, beside what FDA’s draft and Annex 1 say.
| Practice | Typical figure | FDA draft or Annex 1 |
|---|---|---|
| Kit size | 300 to 500 units with 30 to 50 defects, 10% or fewer | FDA: defective units typically no more than 10–20% |
| Kit expiry | Typically 1 year; inspect the kit, then extend | Not addressed. FDA cited a kit procedure with no expiry dates in a 2025 483 |
| Timing | End of the day or shift, for maximum fatigue | Annex 1: worst case, including fatigue |
| Time limit | The routine rate, for example 3 to 4 units a minute | Annex 1: rates controlled and qualified |
| False reject rate | 5% or less | Not addressed |
| First qualification | 3 consecutive passes of the kit | Not addressed |
| Requalification | At least yearly, one pass; also after 3 to 6 months away | Annex 1: at least annually |
| Scope | Separate per product family and container | Annex 1 lists container size among worst cases |
| Training | Typically two weeks | FDA: adequately trained |
FDA’s 483s show where qualification goes wrong. At STAQ Pharma of Ohio, an outsourcing facility, one challenge set served for both training and qualification, and operators did not have to tell critical from major defects, so “operators may fail to identify up to (b)(4) critical defects and still be qualified for visual inspection”.⁷ At UBI Pharma in Taiwan, kit vials were numbered in a way that could identify the defective ones, and inspectors were qualified on a single inspection report.⁶
1.2Setting the pass mark.
No regulation fixes the detection rate. FDA’s recent warning letters say what it expects:
“At a minimum, personnel should reliably be capable of detecting particulates of 150 microns or greater at a 70% probability of detection.” Par Health USA, LLC and Endo USA, Inc., warning letter 320-26-68, 15 April 2026⁸
The same letter records Knapp studies in which personnel “were often not able to achieve a probability of detection of 70%” for critical particles larger than 150 microns.⁸ In August 2026 FDA told Fresenius Medical Care that qualification kits “should typically contain particles in the range of 100-150 microns to adequately challenge inspector detection capabilities.”⁹ In September 2026 it faulted Baxter for challenge kits whose particulates were “excessively large”.¹⁰ Probability of detection comes from a Knapp study, explained in the questions below; AI visual inspection works through one.
2Example: a completed inspector qualification.
| Field | Entry |
|---|---|
| Inspector and scope | VI-017; manual booth inspection of 2 mL clear glass vials of solution products |
| Eyesight | Near vision and colour vision checks passed with usual glasses on 2 November 2026 |
| Test set | QS-2R-04, approved by QA, expires March 2027: 300 units, 270 good and 30 defective. 30 ÷ 300 = 10.0% defective. Identities masked |
| Defects | 8 critical container defects; 18 reject-zone units with extrinsic particles of 100 to 150 microns, so critical; 4 grey-zone units, detected 30–70% of the time by reference inspectors |
| Conditions | Last hour of an 8-hour shift; 2,500 lux at the viewing point, checked before the start; time limit 75 minutes, the routine rate of 4 units a minute |
| Criteria (VIQ-03) | Each class at or above the reference inspectors’ baseline: container defects 90%, reject-zone particles 80%; false rejects no more than 5% of good units; 3 consecutive passes |
| Results, run 3 | Container defects 8 of 8 = 100%. Reject zone 16 of 18 = 88.9%. Grey zone 2 of 4, recorded for trending only. False rejects 4 of 270 = 1.5%. Finished in 71 minutes |
| Decision | Third consecutive pass: qualified for 12 months for this scope. The two missed particle units were reviewed with the inspector. Amber vials and syringes need their own sets |
The criteria come from the reference inspectors’ results for each class, the USP <1790> approach, not a flat 100% for critical defects. Two GSK specialists argued in a 2020 PDA Letter article that a 100% rule tempts sites to fill test sets with obvious defects, so the qualification says little about critical defects that are genuinely hard to see.¹¹
3Defect classification template.
Annex 1 asks for defect classes set during qualification from risk and historical knowledge, including the patient impact and the route of administration.² Record each defect type, not only its class:
| Field | What it holds | Source |
|---|---|---|
| Defect type | One entry per type, such as glass, fibre, metal, crack or missing closure | Catalent letter, 2025¹² |
| Origin | Intrinsic or extrinsic | FDA draft V.B and V.E¹ |
| Class and basis | Critical, major or minor, with the patient risk and route behind it; extrinsic particles critical | Annex 1 8.30;² SCA letter, 2026¹³ |
| Limits | An in-process limit, plus a trend limit from history, often mean + 3σ, reviewed at least yearly | FDA draft V.E;¹ Annex 1 8.33;² slide 37³ |
| Library units | Reference units, images and boundary examples used in training and test sets | Annex 1 8.30;² FDA draft V.C¹ |
| Escalation | Action when found in the AQL sample, a complaint or one container with several particles | FDA draft V.B and V.E;¹ Annex 1 8.30² |
4AQL sampling after 100% inspection.
After the 100% inspection, “the quality unit should sample each batch for acceptance quality limit (AQL) testing.”¹ Part 211 requires written sampling plans with statistical acceptance or rejection levels.¹⁴ FDA’s draft asks for statistically sound plans, operating characteristic curves for each defect class, and a quantified AQL and unacceptable quality limit. It names ASTM E2234, which it says is equivalent to the ANSI/ASQ Z1.4 standards that USP <790> and <1790> reference, and says a more stringent plan may be appropriate for higher-risk products.¹ Annex 1 never mentions AQL, but any critical defect found in later sampling “should trigger an investigation as it indicates a possible failure of the original inspection process.”²
USP <790> requires injections to be “essentially free” of visible particulates, meaning “no more than the specified number of units may be observed to contain visible particulates”.¹⁵ Shabushnig’s slides summarise <790> at batch release as 100% inspection followed by Z1.4 or ISO 2859 sampling at an AQL of 0.65%, with an unacceptable quality limit of 2.3 to 3.3% typical, and give industry mean AQLs of 0.065%, 0.65% and 2.5% for critical, major and minor defects. For product already in distribution, such as units pulled from the market, they give <790>’s plan as 20 units with an accept number of zero: AQL 0.26%, unacceptable quality limit 10.9%.³ Check those against your own copy of USP.
| Lot size | Code, sample | Major, AQL 0.65 | Passes 10% of the time at | Minor, AQL 2.5 | Accept-zero plan |
|---|---|---|---|---|---|
| 501–1,200 | J, 80 | 1 / 2 | 4.78% | 5 / 6 | AQL 0.15 |
| 1,201–3,200 | K, 125 | 2 / 3 | 4.20% | 7 / 8 | AQL 0.10 |
| 3,201–10,000 | L, 200 | 3 / 4 | 3.31% | 10 / 11 | AQL 0.065 |
| 10,001–35,000 | M, 315 | 5 / 6 | 2.92% | 14 / 15 | AQL 0.040 |
| 35,001–150,000 | N, 500 | 7 / 8 | 2.34% | 21 / 22 | AQL 0.025 |
An AQL is the defect rate a plan accepts most of the time, not the rate you release. Read the fourth column: a lot of 1,000 vials with 4.78% major defects still passes 1 time in 10. That is the unacceptable quality limit FDA asks you to quantify, and it is the protection the plan actually gives.¹ For codes L to N it falls between 2.34% and 3.31%, the range the FDA-hosted slides call typical.³
4.1Products that are hard to inspect.
Lyophilised cakes, suspensions, tinted or opaque containers and overwrapped bags hide particles. FDA’s draft suggests advanced technologies such as X-ray, or supplemental destructive testing after the 100% inspection, which may not be warranted if the inspection technology is validated to meet or surpass human inspection.¹ USP <790>, as the slides quote it, requires such batches to be supplemented by inspecting constituted or withdrawn contents of a sample, smaller than a normal AQL sample because the test destroys it.³ In April 2026 FDA told Par Health its remediation plan should include, for products that are difficult to inspect, “implementing destructive testing or other advanced technologies (e.g., (b)(4)) as appropriate, with statistically significant sample sizes to test for critical defects.”⁸
5Example: the AQL sample from a 24,000-vial batch.
Batch 26-311 has 24,000 vials accepted by 100% inspection. Code M gives a sample of 315, drawn across the whole filling run. Critical defects use the accept-zero plan for code M.
| Class | AQL | Accept / reject | Found | Result |
|---|---|---|---|---|
| Critical | 0.040% | 0 / 1 | 0 | Pass |
| Major | 0.65% | 5 / 6 | 4 | Pass |
| Minor | 2.5% | 14 / 15 | 9 | Pass |
The major plan accepts a lot with 0.65% major defects 98.2% of the time, but one with 2.5% only 19.9% of the time. One critical defect in the sample fails the batch whatever the major and minor counts.
Compare the count with the accept number, never a percentage of the batch. Four major defects are 4 ÷ 315 = 1.3% of the sample, but dividing by 24,000 gives 0.017%, which hides the result. FDA’s January 2025 inspection of Turbare Manufacturing, an outsourcing facility, recorded an AQL limit calculated against the batch instead of the sample, and a batch passed without investigation.¹⁷
6When the AQL sample fails.
- Hold the batch and investigate. Identify the particles as intrinsic or extrinsic, and extend the investigation to other batches that may be affected.¹
- Treat a critical defect in the sample as a possible failure of the 100% inspection.²
- Reinspect only with scientific justification, under an approved procedure, with tightened acceptance criteria. “FDA does not recommend more than one reinspection in an attempt to release a batch with atypical defect levels.”¹
- Count units that fail the AQL reinspection with every other reject when reconciling the batch.¹
- Trend defect types and reject levels, and assess product on the market when the trend is adverse.²
- Add any new particle type to the defect library and the test sets, after finding its source.¹
Extrinsic particles found in the 100% inspection or the AQL sample can mean the product is adulterated even when the sampling criteria are met.¹
7What inspectors cite.
Amneal’s polypropylene bags shed white fibres, which its investigations had traced since late 2022. Based on those investigations it relaxed its limits for major defects and total rejects, kept using the bags, and after a 2024 investigation released two batches of ropivacaine infusion bags, which can be given into the epidural space, after reinspection and a risk assessment FDA found inadequate:¹⁸
“It is unacceptable to increase lot acceptance specifications to excessive levels to enable use of unsuitable polypropylene bags received from your supplier.” Amneal Pharmaceuticals, LLC, warning letter 320-25-104, 27 August 2025¹⁸
“Extrinsic particulates should be classified as critical, and the intrinsic particulates should be evaluated based on multiple factors, including particle size and route of administration, to determine the appropriate defect classification.” SCA Pharmaceuticals, LLC, warning letter, 14 September 2026¹³
“Operators’ rate of inspection was excessively fast (e.g., (b)(4) per unit) and did not align with the standards referenced in your procedures.” Baxter International Inc., warning letter 320-26-132, 29 September 2026¹⁰
Par Health classified critical defects as major, so batches passed AQL testing that failed once reclassified, and its procedures allowed several reinspections.⁸ Catalent Indiana set limits only by class, not by defect type.¹²
8What changed from 2024 to 2026.
- FDA guidance. The December 2021 draft on visible particulates was still a draft when this page was reviewed. Its advice against more than one reinspection reappears in the April 2026 Par letter.¹⁸
- Annex 1. Annual qualification, worst-case conditions, breaks and statistical trending of defect types have applied in the EU since 25 August 2023.² PIC/S PE 009 carries the same text.¹⁹ In the UK, where regulation C17 keeps the EU guidelines as they stood at the end of 2020 unless the MHRA replaces them, the MHRA Orange Guide’s March 2023 update carries the revised Annex 1 from the same date.²⁰²¹
- Enforcement. The detection minimum (Par), kit particle sizes (Fresenius and Baxter) and the qualification faults (UBI and STAQ) on this page come from documents issued between October 2025 and September 2026.⁸⁹¹⁰⁶⁷
9How it runs in Seal.
In Seal, visual inspection and defect management holds the defect library with its boundary examples, look-alikes and effective dates, and each inspector qualification records true positives, false negatives, false positives and classification accuracy against a controlled challenge set. Qualification can differ by product family, container, defect class and method, so a person qualified on clear vials is not assumed qualified for amber syringes. First-pass findings stay in the record through reinspection, and the AQL result is kept distinct from the 100% inspection while both are assembled for disposition.
The same qualification records can gate the work itself: training and qualification lets you configure the prerequisite for a task and verify both the permitted and the blocked path, so prerequisites such as an annual qualification and an eyesight check become a condition of starting the inspection.
References
- 1Inspection of Injectable Products for Visible Particulates, draft guidance for industry, sections V, V.A.2 to V.A.4, V.B, V.C and V.E, full text searched for “expir”, “false reject” and “consecutive”, FDA, December 2021. https:/
/ www. fda. gov/ media/ 154868/ download (read 8 November 2026) - 2EudraLex Volume 4, Annex 1: Manufacture of Sterile Medicinal Products, sections 8.30 to 8.33, full text searched for “AQL”, “expir”, “false reject” and “consecutive”, European Commission, 22 August 2022, in operation since 25 August 2023. https:/
/ health. ec. europa. eu/ system/ files/ 2022- 08/ 20220825_ gmp- an1_ en_ 0. pdf (read 8 November 2026) - 3Shabushnig JG, Visual Inspection of Injectable Products: More than Sorting Good from Bad, slides 9, 12 to 14 and 31 to 38 on USP <790>, Ph. Eur. 2.9.20, JP 6.06, training, qualification, in-process limits and acceptance sampling, Insight Pharma Consulting, September 2022, hosted by FDA. https:/
/ www. fda. gov/ media/ 162175/ download (read 8 November 2026) - 4Melchore JA, Sound Practices for Consistent Human Visual Inspection, AAPS PharmSciTech 12(1):215–221, 2011, the paper cited in FDA’s draft guidance as Melchore 2011. https:/
/ pmc. ncbi. nlm. nih. gov/ articles/ PMC3066354/ (read 8 November 2026) - 5Revision of the Ph. Eur. dosage form monograph on Parenteral preparations (0520) adopted, EDQM newsroom, 17 September 2020. https:/
/ www. edqm. eu/ en/ newslist/ - / asset_ publisher/ JaJ4OiL8a8pm/ content/ id/ 95277 (read 8 November 2026) - 6Form FDA 483, UBI Pharma Inc., Hsinchu, Taiwan, inspection 6 to 14 October 2025, observation 2, FDA, issued 14 October 2025. https:/
/ www. fda. gov/ media/ 190241/ download (read 8 November 2026) - 7Form FDA 483, STAQ Pharma of Ohio, LLC, Columbus, Ohio, outsourcing facility, inspection 10 to 19 August 2026, observation 1, FDA, issued 19 August 2026. https:/
/ www. fda. gov/ media/ 194821/ download (read 8 November 2026) - 8Par Health USA, LLC and Endo USA, Inc., warning letter 320-26-68, FDA, 15 April 2026. https:/
/ www. fda. gov/ inspections- compliance- enforcement- and- criminal- investigations/ warning- letters/ par- health- usa- llc- endo- usa- inc- 722121- 04152026 (read 8 November 2026) - 9Fresenius Medical Care AG & Co. KGaA (Ogden plant), warning letter 320-26-119, FDA, 25 August 2026. https:/
/ www. fda. gov/ inspections- compliance- enforcement- and- criminal- investigations/ warning- letters/ fresenius- medical- care- ag- co- kgaa- 730319- 08252026 (read 8 November 2026) - 10Baxter International Inc., warning letter 320-26-132, MARCS-CMS 733231, FDA, 29 September 2026. https:/
/ www. fda. gov/ inspections- compliance- enforcement- and- criminal- investigations/ warning- letters/ baxter- international- inc- 733231- 09292026 (read 8 November 2026) - 11Flaquiere A and Malthête J, Qualification of Manual Visual Inspection Still Critical, PDA Letter, Parenteral Drug Association, 19 March 2020. https:/
/ www. pda. org/ pda- letter- portal/ home/ full- article/ qualification- of- manual- visual- inspection- still- critical (read 8 November 2026) - 12Catalent Indiana, LLC, warning letter 320-26-20, FDA, 20 November 2025. https:/
/ www. fda. gov/ inspections- compliance- enforcement- and- criminal- investigations/ warning- letters/ catalent- indiana- llc- 718189- 11202025 (read 8 November 2026) - 13SCA Pharmaceuticals, LLC, outsourcing facility, warning letter, MARCS-CMS 724202, FDA, 14 September 2026. https:/
/ www. fda. gov/ inspections- compliance- enforcement- and- criminal- investigations/ warning- letters/ sca- pharmaceuticals- llc- 724202- 09142026 (read 8 November 2026) - 1421 CFR 211.165, Testing and release for distribution, paragraphs (c) and (d), eCFR. https:/
/ www. ecfr. gov/ current/ title- 21/ section- 211. 165 (read 8 November 2026) - 15USP General Chapter <790> Visible Particulates in Injections, public summary, United States Pharmacopeia, 2017. https:/
/ doi. org/ 10. 31003/ USPNF_ M7197_ 01_ 01 (read 8 November 2026) - 16MIL-STD-105E, Sampling Procedures and Tables for Inspection by Attributes, Tables I and II-A, US Department of Defense, 10 May 1989, and Notice 1 cancelling it in favour of ANSI/ASQC Z1.4-1993, 27 February 1995. https:/
/ everyspec. com/ MIL- STD/ MIL- STD- 0100- 0299/ MIL- STD- 105E_ 11675/ (read 8 November 2026) - 17Form FDA 483, Turbare Manufacturing, Conway, Arkansas, inspection 14 to 24 January 2025, FDA. https:/
/ www. fda. gov/ media/ 185399/ download (read 8 November 2026) - 18Amneal Pharmaceuticals, LLC, warning letter 320-25-104, on its Amneal Pharmaceuticals Private Limited facility in Gujarat, India, FDA, 27 August 2025. https:/
/ www. fda. gov/ inspections- compliance- enforcement- and- criminal- investigations/ warning- letters/ amneal- pharmaceuticals- llc- 709894- 08272025 (read 8 November 2026) - 19PE 009-18 (Annexes), Guide to Good Manufacturing Practice for Medicinal Products, Annex 1, sections 8.30 to 8.33, PIC/S, 24 September 2026. https:/
/ picscheme. org/ docview/ 11333 (read 8 November 2026) - 20The Human Medicines Regulations 2012 (SI 2012/1916), regulation C17, legislation.gov.uk, up to date to 7 November 2026. https:/
/ www. legislation. gov. uk/ uksi/ 2012/ 1916/ regulation/ C17 (read 8 November 2026) - 21The MHRA Orange Guide March 2023 Update, Pharmaceutical Press, publisher of the MHRA Orange Guide. https:/
/ www. pharmaceuticalpress. com/ ?p=2808 (read 8 November 2026)
AQuestions and answers
What is a Knapp test?
A method, published by Knapp and Kushner in 1980, for measuring how well an inspection finds defects. Reference inspectors inspect each unit of a test set many times, giving every unit a probability of detection. Units rejected more than 70% of the time form the reject zone, and a new inspector or an automated machine is compared with the manual result on those units.
Which inspection level is used for AQL sampling of injectables?
General inspection level II is the usual choice. MIL-STD-105E, the military standard that ANSI/ASQ Z1.4 replaced, says level II is normally used, level I gives less discrimination and level III more. Its special levels S-1 to S-4 keep samples small and accept larger sampling risks. FDA’s draft guidance adds that higher-risk products may need a more stringent sampling plan.
How often must visual inspectors be requalified?
At least annually under EU GMP Annex 1. FDA’s draft guidance leaves the frequency to the qualification protocol, which the quality unit approves. Set other triggers in the protocol too, such as a failed qualification, a long absence, or a change to the inspection conditions, the container or the defect library.
Can an inspector who wears glasses be qualified?
Yes. Annex 1 expects the qualification to be done wearing corrective lenses if the inspector normally wears them, and FDA’s draft guidance asks for normal near visual acuity with or without corrective lenses. Colour vision should not be impaired.
What is the difference between USP <790> and USP <1790>?
USP <790> sets the requirement that injections be essentially free of visible particulates and defines that term by the number of units allowed to contain them. USP <1790> is guidance on inspecting injections for visible particles, including inspector qualification and sampling plans. It treats visual inspection as a probabilistic process.