The healthy volunteer effect was studied by comparing 6 years of mortality data for 31,124 participants from the Adventist Health Study (AHS) who responded to both a relatively brief census questionnaire (CQ) in 1974 and a detailed life-style questionnaire (LQ) in 1976 (responders), to mortality data for 8,762 individuals who did not respond to the second questionnaire. The rate ratio (RR) comparing LQ nonresponders to responders for all cause mortality decreased from 2.5 (2.2-2.9) in 1977 to 1.4 (1.2-1.7) in 1982 (p for trend = 0.02); for ischemic heart disease mortality from 2.3 (1.8-3.0) to 1.3 (1.0-1.7); and for all sites cancer mortality from 1.8 (1.3-2.5) to 1.5 (1.1-2.0). The death rate decreased markedly among nonresponders and increased slightly among responders during the study. Similar results were seen for age and gender subgroups. Multivariate analysis controlling for confounding variables confirms these results, except that the apparent effect of education is probably due to effect modification by age. The RR decreased to about one after 3 years of follow-up in young subjects but remained elevated (> 2) in older subjects. Available sociodemographic information reveals that a higher proportion of responders are married, have college education, are SDA church members, and use medical services less than nonresponders during the previous year. Because the risk remains elevated at the end of the study in some but not all subgroups, it seems reasonable that the elevated risk in nonresponders may be due in part to a less healthy life style and in part to exclusion of individuals who did not feel well during enrollment. The results suggest that for internal comparisons no bias is likely to occur; but descriptive statistics for certain subgroup comparisons, and external comparisons, may be biased by the healthy volunteer effect.