The Behavioral Risk Factor Survey (BRFS) is a telephone interview used widely by the Centers for Disease Control (CDC) in conjunction with state health departments to measure prevalences and time trends for health-related behaviors. We assessed the reliability of those parts of the BRFS related to cardiovascular disease (smoking, diet, obesity, exercise, and hypertension) and demographics by readministering the BRFS in July and August of 1989 to 145 randomly selected subjects between 10 and 21 days following completion of an initial interview. Sampling and data collection closely followed CDC procedures. The retest samples comprised 49 whites, 43 blacks, and 53 Hispanics living in northern Manhattan in New York City. Group prevalences or means were compared at first and second interviews for six demographic variables and 12 behavioral risk factor variables for the samples as a whole and separately for each ethnic group. All of these comparisons were highly consistent, and none showed a statistically significant difference. At the individual level, Pearson or kappa correlations for 19 questions related to demographics and behavioral risk factors other than diet were greater than or equal to 0.60 (p less than 0.001) for all except routine checkup in the past 2 years (kappa = 0.54; p less than 0.001) and blood pressure measured in the past 2 years (crude concordance, 96%; kappa = 0.23; p less than 0.01). For 17 food items, correlations for frequency of consumption ranged from 0.44 to 0.76 (p less than 0.01). For a composite index of diet "atherogenicity" based on the 17 food items, r = 0.62 (p less than 0.001). This test-retest reliability study of the BRFS showed high consistency at the group level and acceptable to high item reliability at the individual level for the parts of the BRFS related to demographics, cardiovascular behavioral risk factors, and a 17-item nutrition module. Findings were generally consistent in all three ethnic groups.