ABSTRACT CONTEXT AND OBJECTIVE: The food frequency questionnaire (FFQ) is the preferred instrument for obtaining dietary information in epidemiological studies. A short form of the FFQ was compared with the original version that was used in the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil), and also with three 24-hour dietary recalls. DESIGN AND SETTING: Cross-sectional study carried out in six Brazilian state capitals. METHODS: Multiple linear regression was used to reduce the original food and drink list of the FFQ, which had contained 114 food items. The frequency of consumption and nutritional composition of the foods were also taken into consideration. To assess the validity of the shortened FFQ, the energy and nutrients values of the 24-hour dietary recalls were deattenuated and log-transformed. RESULTS: The list of the FFQ of ELSA-Brasil was reduced to 76 food items. The intraclass correlation coefficients in the validation study ranged from 0.17 (selenium) to 0.66 (calcium). CONCLUSIONS: The number of items was reduced by 33%, while still maintaining relatively good capacity to measure energy and selected nutrients.