Background  Food frequency questionnaires (FFQs) are widely used in nutritional epidemiology but no papers detail the development of the supporting programs and nutritional databases. The principles involved in data collection, processing and treatment of the European Prospective Investigation into Cancer Study (EPIC)‐Norfolk Study FFQ and development of the structure and content of the Compositional Analyses from Frequency Estimates (CAFE) program for calculating nutrient intakes are described. Extreme nutrient values and derivation of cut‐points for data exclusion are also discussed.

Methods  The FFQ was sent to 25 637 participants. To exclude extreme (nonphysiological) outliers of nutrient intakes, the extreme top and bottom 0.5% of the ratio of energy intake to estimated basal metabolic rate was used.

Results  The CAFE computer program modifies nutrient intake according to specific fats used in food preparation and the amount of visible fat on meat. It incorporates different breakfast cereals and distinguishes between text for brands and types. After exclusion of outliers of nutrient intake some extreme values remained for energy, carbohydrate, protein, fat and alcohol, because of high reported frequencies of certain foods.

Conclusions  The features of CAFE include flexible, updateable, databases and a novel method for matching text. The effect of extreme nutrient values on the accuracy of the FFQ in estimating diet in nutritional epidemiology should be investigated further.