To aid in the search for more practical and reliable methods for use in population-based studies of diabetes mellitus, this article compares five ways of estimating prevalence rates. The analysis was performed on secondary data from a cross-sectional study of a cluster sample of the adult population in nine state capitals in Brazil. The original study was carried out from 1986 to 1988. The 21,846 participants were classified as diabetic or not diabetic by five different methods: household questionnaires administered to the entire sample population (M1); individual questionnaires administered to the selected population (M2); measurement of fasting glucose levels in capillary blood, with levels > or = 120 mg/dL as the cutoff (M3); individual questionnaire and fasting capillary blood glucose > or = 120 mg/dL (M4); and individual questionnaire plus fasting capillary blood glucose > or = 120 mg/dL and capillary glucose 2 hours after oral glucose loading > or = 200 mg/dL (M5). Agreement between the methods was determined by comparison of the rates obtained and use of the kappa coefficient. The age-adjusted prevalence rates of diabetes varied according to the method used. Values obtained with M1 were lower than those indicated by M2; M3 values were higher than M2 values, except in the age group 60-69 years; and with M5 the rates were higher than with M4, except among persons 30-39 years old. With regard to the age-adjusted rates found by the various methods, M1 detected 84% of the M2 estimate, M2 detected 91% of the M3, M3 detected 70% of the M4, and M4 detected 86% of the M5. Previously diagnosed diabetes cases accounted for 64% and 55% of the totals estimated by M4 and M5, respectively. Kappa values were at least 0.70 for M1 compared against M2, M1 against M4, M2 against M4, and M3 against M4. Based on the results of this study, it was concluded that the questionnaires used in M1 and M2 constituted appropriate methods for detecting previously diagnosed cases of diabetes mellitus, and their use is recommended for the purposes of health services planning or evaluation. Fasting glucose measurement (M3) as the sole method did not show a significant advantage over the individual questionnaire (M2). Of the combined or multiple methods, fasting glucose together with the individual questionnaire (M4) was efficient in comparison to M5, which incorporated measurement of blood glucose 2 hours after oral glucose ingestion.