Foods to be included in a Danish self-administered semiquantitative food frequency questionnaire were identified from food tables developed, together with data collected, for the survey 'Dietary habits in Denmark, 1985'. The questionnaire was to be used in a prospective study on diet, cancer and health, and the aim was to rank individuals with regard to intake of 19 different nutrients considered of prime importance in human carcinogenesis. The questionnaire for the dietary survey included 247 foods and recipes. From stepwise multiple regression analyses with the intake of each of the 19 nutrients as the dependent variable and the intake of the 247 foods and recipes as independent variables, the foods in the models explaining 90% of the between-person variability were considered for the final questionnaire. All relevant analyses were performed for the study group as a whole, for men and women separately, and in each gender for subgroups of energy intake. Taken together, the models explaining 90% of the between-person variability identified a total of 74 foods or recipes, which were important predictors of the intake of one or more of the nutrients considered. A few foods were excluded and a few foods were added to the final questionnaire based on common biological background information, and on information on foods providing important amounts of given nutrients, but which failed to contribute to regression analyses. The 92 foods and recipes, which were included in the final questionnaire provided altogether 81% of the average total supply of the nutrients.(ABSTRACT TRUNCATED AT 250 WORDS)