Although self-administered questionnaires are major sources of information in epidemiology, comparatively little has been done to study practical aspects of design and mailing. The objective of this study was to evaluate various measures taken to increase the response rate. A questionnaire was mailed in July 1995 to a random sample (n = 2,000) of the Swedish population aged 20-79 years. Using a randomized factorial study design, the questionnaire and mailing procedures were changed in three ways: preliminary notification, length of the questionnaire, and mention of telephone contact. The overall questionnaire retrieval rate was 49%. Preliminary notification (adjusted odds ratio of receiving a completed questionnaire = 1.30, 95% confidence interval (CI) 1.08-1.56 relative to the absence of preliminary notification) and short length of the questionnaire (odds ratio = 1.24, 95% CI 1.04-1.48 relative to a long questionnaire) were both independently associated with a higher retrieval rate. Of eight possible combinations, the one comprising preliminary notification, a short questionnaire, and no mention of telephone contact gave the highest retrieval rate, 56%. The lowest retrieval rate, 40%, was observed for the combination of no preliminary notification, a long questionnaire, and mention of telephone contact. Young age, male sex, and urban residence significantly lowered the retrieval rate. Although there was a positive association between the questionnaire retrieval rate and partial nonresponse (missing answers in retrieved questionnaires), the marginal losses due to the latter did not cancel the gains by optimized mailing routines. Old age was the strongest determinant of partial nonresponse. The data provide evidence that design and mailing strategies, as well as demographic characteristics, may greatly influence the response rate of mailed epidemiologic questionnaires.