This study was conducted to evaluate the agreement between retrospective information on cigarette smoking and original information obtained six years previously. A questionnaire on smoking habits was answered by 94% of a random sample of the adult Swedish population in 1963 (original information). Of the 10,356 subjects aged 40-59 years who reported that they had ever smoked regularly, 9,394 answered a second questionnaire in 1969. Information was obtained on present smoking habits as well as on smoking habits in 1963 (retrospective information). Cigarette consumption was divided into four categories: 0, 1-7, 8-15, and 16+ cigarettes per day. In relation to the original information on smoking, retrospective information showed a strong tendency to overestimate previous cigarette consumption among subjects who had increased their cigarette smoking (69%), and to underestimate previous cigarette consumption among subjects who had decreased their cigarette smoking (49%). Subjects with unchanged cigarette consumption showed a high level of agreement between original and retrospective information (89%). Age, sex, income, and occupational status also had some impact on the agreement between retrospective and original information on cigarette consumption. Consequences for epidemiologic studies based on retrospective smoking information are discussed. The use of retrospective information may introduce considerable differential misclassification when cases and noncases differ with regard to changes in smoking habits.