Survey (FNDDS)

FNDDS, the Food and Nutrient Database for Dietary Studies, is the USDA Agricultural Research Service's purpose-built database for coding 24-hour dietary recalls. It is the dataset that sits between the words an NHANES respondent says into a phone — 'a slice of cheese pizza, thin crust, about the size of my hand' — and a table of kilocalories, grams of saturated fat, and milligrams of sodium attributed to that respondent. Every major U.S. dietary-epidemiology finding of the last two decades is, mechanically, a FNDDS product.

How FNDDS is built

FNDDS is published on a two-year cycle aligned to the NHANES survey waves — FNDDS 2019-2020, FNDDS 2021-2022, and so on. Each release contains roughly 7,000 to 8,000 foods coded into the USDA's What We Eat in America (WWEIA) food category system, plus the portion-size estimation references used by trained interviewers during the Automated Multiple-Pass Method dietary recall. The nutrient values are not original analyses — they are inherited from SR Legacy and Foundation Foods and propagated through recipe-style modelled entries for mixed dishes.

The recipe layer

The non-trivial work inside FNDDS is the recipe modelling for complex foods. A 'cheese pizza, thin crust, one slice' FNDDS entry is not an analysed sample; it is a modelled composite built from ingredient proportions (typical crust, typical sauce, typical cheese) multiplied by ingredient nutrient profiles from Foundation Foods and SR Legacy, adjusted for cooking yield and moisture loss. This modelling is explicit and documented, which is a strength — but it means every mixed-dish figure in NHANES output is subject to recipe-composition assumptions as well as nutrient-analysis assumptions.

Why consumer apps rarely use FNDDS directly

Consumer tracking apps overwhelmingly draw from SR Legacy and Branded Foods rather than FNDDS because FNDDS entries are calibrated for recall coding — 'pizza, thin crust' — rather than for the label-style specificity a user logging their own meal wants ('Domino's large hand-tossed pepperoni'). The specificity is backwards. FNDDS is optimised for epidemiology; Branded Foods is optimised for labelling. Each has its place, and confusing them produces dietary-assessment figures that are technically consistent and practically meaningless.

Known weaknesses

FNDDS is only as current as its inheritance. When Foundation Foods has not re-analysed a commodity, FNDDS carries forward SR Legacy figures, sometimes decades old. Recipe models reflect mid-2010s American preparation norms and lag behind real-world shifts — vegan cheese, plant-based burgers, and ultra-processed analogues of traditional foods enter FNDDS with a lag of roughly one release cycle after they become common. Published critiques in the American Journal of Clinical Nutrition have flagged this lag as a nontrivial source of bias in dietary-change estimation.