Validity of recalled cigarette data was assessed among 87 middle-aged adults by comparing recall to longitudinal records. Agreement on smoking status and amount smoked 20 years ago occurred for 87 per cent and 71 per cent of subjects, respectively. Corresponding proportions for 32-year recall were 84 per cent and 55 per cent. Gender did not influence recall accuracy. Ex-smokers tended to make the most errors. Accuracy of recalled smoking information 20 years ago was comparable to that of alcohol status and consumption frequency category.