When a Food Recall Crosses the Tape, Its Class Is Usually Still Pending

The number that sizes a food recall is not the case count, the number of states listed, or the prominence of the headline. It is the classification — and on the morning a recall notice first appears, that classification frequently has not been assigned yet.

Recall classification is a formal administrative act with a published definition, performed by a federal agency, on a schedule slower than the press release. Two notices carrying the same grade can differ by three orders of magnitude in the quantity of product that has to be retrieved. Reading one properly means separating three things that arrive at different times: what the firm said, how the agency graded it, and how deep into the distribution chain the retrieval goes.

The grade has a written definition, and it is not about size

The Food and Drug Administration's classifications are defined in 21 CFR 7.3(m). The regulation calls the classification "the numerical designation, i.e., I, II, or III, assigned by the Food and Drug Administration to a particular product recall." The three tiers read as follows.

• Class I — "a situation in which there is a reasonable probability that the use of, or exposure to, a violative product will cause serious adverse health consequences" 7.3(m)(1)
• Class II — "a situation in which use of, or exposure to, a violative product may cause temporary or medically reversible adverse health consequences" 7.3(m)(2)
• Class III — "a situation in which use of, or exposure to, a violative product is not likely to cause adverse health consequences" 7.3(m)(3)

Every one of those definitions is a statement about the probability and reversibility of harm to a person who eats the product. None of them is a statement about how many units were made, how many retail chains carried them, or how large the recalling firm is. A Class I designation on a single lot from a single plant and a Class I designation on a nationally distributed ingredient are the same grade describing very different operational problems.

The regulation also names the inputs to the grading decision. Under 21 CFR 7.41(a), the health hazard evaluation considers whether "any disease or injuries have already occurred from the use of the product," the hazard to "those individuals who may be at greatest risk," and the seriousness, likelihood and consequences of the hazard. The first factor carries the most weight in practice: a recall issued after illnesses have already been reported is a different situation from one issued after a routine environmental swab, even when both end up graded the same way.

Two agencies share the word, and they do not share the rulebook

A second sorting step comes before the grade. Food recalls in the United States are administered by two separate agencies with separate statutes, separate publication channels, and different enforcement powers. Which agency issued the notice determines which rulebook applies to everything that follows.

Figure 1. Jurisdiction and enforcement powers. Sources: 21 CFR Part 7; 21 U.S.C. 350l; FSIS Directive 8080.1 Rev 8; Federal Meat Inspection Act 21 U.S.C. 601, Poultry Products Inspection Act 21 U.S.C. 451, Egg Products Inspection Act 21 U.S.C. 1031.

The Food Safety and Inspection Service, inside USDA, administers the Federal Meat Inspection Act, the Poultry Products Inspection Act, and the Egg Products Inspection Act. FDA covers essentially everything else in the food supply. FSIS uses the same Roman numerals with slightly different wording — its Class I is a "health hazard situation where there is a reasonable probability that the use of the product will cause serious, adverse health consequences or death," its Class II covers a "remote probability of adverse health consequences," and its Class III covers situations where the risk is negligible.

The wording is close enough that the two systems are often treated as interchangeable. The enforcement mechanics are not. FSIS states plainly that the recalls it coordinates are initiated voluntarily by the company; if a firm refuses, FSIS can detain the product and ask the Department of Justice to pursue seizure, but it cannot order the recall itself. FDA does have that power, and the conditions are narrow. Under 21 U.S.C. 350l(a), the agency acts when it determines "there is a reasonable probability that an article of food (other than infant formula) is adulterated under section 342 of this title or misbranded under section 343(w) of this title and the use of or exposure to such article will cause serious adverse health consequences or death," and even then it "shall provide the responsible party ... with an opportunity to cease distribution and recall such article" first. Only if the party refuses may FDA order distribution to stop, and the statute provides for an informal hearing not later than two days after the order issues.

How often that machinery runs is measurable. Across the full openFDA food enforcement archive, 28,907 records are marked "Voluntary: Firm initiated" against 396 marked "FDA Mandated" — roughly 1.4 percent of 29,310 total records. In the 2025 report year, every one of the 1,617 records carries the voluntary flag. A recall is, by default, something a company chooses to do, usually after a conversation with an agency it would prefer not to escalate with.

The announcement is step three of five

The sequence matters more than any single document in it. A firm that determines an article of food is a "reportable food" — defined in 21 U.S.C. 350f(a)(2) as one "for which there is a reasonable probability that the use of, or exposure to, such article of food will cause serious adverse health consequences or death to humans or animals" — must submit a report through the Reportable Food Registry "as soon as practicable, but in no case later than 24 hours after" making that determination, under 350f(d)(1). That filing is not public. The press release that follows it is.

Figure 2. Sequence and depth. Sources: 21 U.S.C. 350f(d)(1); FDA Enforcement Report information and definitions; 21 CFR 7.42(b)(1).

FDA is explicit that the grade trails the announcement. In its own Enforcement Report documentation, the agency notes that "most of the time recall classifications are determined after the recalling firm provides all of the information needed," and that firms often issue recalls and public statements "well before the FDA completes its classification process and subsequently posts to this report." Records awaiting that determination sit in the report as "Not Yet Classified."

The practical consequence is a window. During it, the only characterization of severity in circulation is the one the recalling firm wrote, filtered through whichever outlets picked it up. The federal grade, the depth of recall, and the quantity in distribution channels all land afterward, in the weekly Enforcement Report, in a format that gets a fraction of the attention the original notice received. A position that was sized against the headline was sized against the least informative artifact in the sequence.

Class I is common, which is exactly why it is weak evidence

If the grade were rare, its appearance would carry information on its own. It is not rare.

Figure 3. Source: openFDA Food Enforcement API, api.fda.gov/food/enforcement.json, query count=classification.exact filtered by search=report_date:[YYYY0101+TO+YYYY1231], retrieved August 2026. Counts are enforcement-report records, not distinct recall events.

Across the seven report years shown, Class I records range from 408 in 2021 to 1,224 in 2023, and the Class I share of annual records runs from roughly 35 percent in 2019, 2021 and 2024 to roughly 57 percent in 2023. Class III — the tier the regulation describes as "not likely to cause adverse health consequences" — never exceeds 183 records in a year and accounts for about 6 percent of the 29,310-record archive. Put plainly: the serious tier is the ordinary tier, and the trivial tier is the rare one.

That inverts the intuitive reading. A notice that says "Class I" is not telling a reader that something unusual has happened; it is telling a reader that this notice resembles roughly half of all classified food enforcement records. What separates one Class I from another is everything the class label leaves out.

The second dial: how deep the retrieval goes

That missing information has its own regulation. Under 21 CFR 7.42(b)(1), a recall strategy specifies a depth: consumer or user level, retail level including any intermediate wholesale level, or wholesale level. Those three options describe entirely different operational burdens for the same nominal grade. A wholesale-level recall means the units never reached a shelf. A consumer-level recall means retrieval has to reach households.

The same section calibrates verification. Effectiveness checks under 7.42(b)(3) run from Level A, "100 percent of the total number of consignees to be contacted," through Level B (greater than 10 percent and less than 100 percent), Level C (10 percent), Level D (2 percent), to Level E, "no effectiveness checks." A firm running Level A checks to the consumer level is performing a fundamentally larger exercise than one running Level D checks to the wholesale level, whatever the Roman numeral on the file says.

Quantity has a home too. 21 CFR 7.46(a) lists nine items a firm should provide when it initiates a recall, and items four through six are the ones that convert a notice into a scale estimate: "total amount of such products produced and/or the timespan of the production," "total amount of such products estimated to be in distribution channels," and "distribution information, including the number of direct accounts." Public notification of the classified recall then follows under 7.50, in the weekly Enforcement Report, "according to its classification."

Reading a notice without guessing at the tape

None of this produces a directional call, and treating it as one is the failure mode worth naming. Grade, depth and quantity describe the size of an operational and legal problem. They do not describe what has already been discounted, how much of the affected line contributes to a given issuer's revenue, or whether a supplier three tiers upstream is more exposed than the brand named in the headline. A recall of an ingredient can propagate into a second wave of downstream notices from firms that bought it: FSIS has issued public health alerts covering meat and poultry products that contained FDA-regulated ingredients already under recall.

What the framework does provide is a way to tell, within hours rather than weeks, whether a notice is describing a bounded event or an open-ended one. A single production lot, a single establishment number, a wholesale-level depth and no reported illnesses is a bounded event whose paperwork will conclude. An ingredient shipped to unnamed downstream manufacturers, a consumer-level depth, an open illness investigation and a "Not Yet Classified" status is an event whose full extent is not yet knowable from any public document — which is a legitimate reason to size a position smaller, or not to take one, rather than a reason to guess harder.

What Would Invalidate This

Several conditions break the frame, and they are common enough to check before applying it.

• The grade is already public at announcement. FSIS assigns the class through its Event Assessment Committee as part of recommending the recall, so an FSIS notice does not necessarily have the same classification lag FDA describes for its own process. The exact publication timing of the FSIS class relative to the notice was not verifiable from the agency documents reviewed here, so treat the lag as an FDA-side property until confirmed otherwise for a specific USDA case.
• The record counts are records, not events. Each openFDA food enforcement entry is one enforcement-report line, and a single recall event can generate several entries covering different products or lots. Year-over-year changes in the counts above can therefore reflect how a few large multi-product events were recorded rather than a change in underlying recall frequency.
• Report year is not event year. The chart filters on report_date, which is when the record was reported, not when contamination occurred or when the firm announced. A recall announced in December can land in the following report year.
• Small firms and private firms. Most food recalls involve companies with no listed equity at all. A grade-and-depth read tells you about the recall; it tells you nothing tradable when the recalling entity is a private co-packer whose customers are not disclosed.
• Market withdrawals are not recalls. Under 21 CFR 7.3(j), a market withdrawal is "a firm's removal or correction of a distributed product which involves a minor violation that would not be subject to legal action." These do not receive a class and do not appear in the classified enforcement record, so an absence from the database is not evidence that nothing happened.
• Commercial consequences sit outside the regulation entirely. Retailer delisting, private-label contract loss and insurance treatment follow commercial logic, not the classification scheme.

Concrete Framework

A repeatable sequence for a food recall notice, run in this order and stopped at the first step that returns "not knowable."

1. Identify the agency first. Meat, poultry or processed egg products means FSIS and the FMIA, PPIA or EPIA. Everything else means FDA and 21 CFR Part 7. Do not proceed until this is settled, because the remaining steps differ.
2. Record whether a class is stated. If the notice is FDA-side and carries no class, log the event as unclassified and check the weekly Enforcement Report for the record moving out of "Not Yet Classified." Treat the interval as a period of incomplete information, not as a period of good news.
3. Find the depth. Locate whether the strategy runs to wholesale, retail, or consumer level under 7.42(b)(1). A consumer-level notice with a public warning is a different exercise from a wholesale-level notice with neither.
4. Find the quantity. Look for the figures that map to 7.46(a) items four through six — amount produced, amount estimated to be in distribution channels, number of direct accounts. If none are disclosed, the scale is genuinely unknown, and that is the finding.
5. Check for existing illness reports. This is the first factor in the 7.41(a) hazard evaluation and the single input most likely to drive escalation from a firm-initiated notice to a broader investigation.
6. Trace one tier upstream. Ask whether the recalled item is a finished good or an ingredient. Ingredient-level events generate downstream notices from other firms; finished-good events usually do not.
7. Write the invalidation condition before sizing anything. State in advance what would tell you the read was wrong — a class assignment that lands lower than expected, a depth confined to wholesale, a quantity disclosure that turns out to be a rounding error against the issuer's volumes — and let that condition, not the next headline, govern the exit.
8. Log the outcome against the initial read. Record the announced grade, the final classified grade, the depth, the disclosed quantity, and how long the classification took. A dozen entries of this kind is enough to calibrate how much weight the initial notice deserves relative to the classified record that follows it.