Reliability of recalled measures of physical activity and alcohol and tobacco use was examined in a cohort of 873 men and women in three California communities. In personal interviews in 1972, participants provided baseline data on these three habits. In repeat interviews in 1983, they recalled their habits in 1972 and reported their current habits. On average, recalled physical activity levels significantly exceeded those originally reported (1899 kcal/d versus 1345 kcal/d, P < 0.001) and were similar to current reported levels (1822 kcal/d). Similarly, recalled alcohol and cigarette consumption was higher than the original reports (alcohol: 126 versus 119 g/wk; cigarettes: 8.5 versus 6.2/d). By contrast, current reported alcohol (103 g/wk) and cigarette consumption (4.6/d) were lower than at baseline. Analysis of variance was used to partition the variation in recalled and original habits into components due to interpersonal variation in true measures, errors in recall, and residual reporting error. Interpersonal variation accounted for only 20% of the total variation in physical activity, but for 48 to 60% of total variation in cigarette smoking and alcohol consumption. These results suggest that inconsistencies among studies of chronic disease and physical activity may arise from either large random measurement errors in individual assessments or from homogeneity of activity levels among the populations studied.