To evaluate the magnitude and duration of the "healthy volunteer effect" (HVE) in a middle-aged general population in Japan, we followed 54,396 subjects (26,998 men, 27,398 women) of the JPHC Study Cohort I from 1990 through 1999, and compared cause-specific mortality between respondents (79%) and nonrespondents to a baseline questionnaire survey. Age and area-adjusted RRs of mortality were higher among nonrespondents for all causes (2.02, 95% CI = 1.83-2.24), all cancers (1.43, 1.20-1.71), all circulatory system diseases (2.26, 1.86-2.74), and cerebrovascular disease (2.73, 2.04-3.66) in men, and 1.63 (1.39-1.92), 1.22 (0.94-1.58), 1.53 (1.07-2.20), and 1.65 (1.02-2.65), respectively, in women. These effects were not observed for ischemic heart disease. RR elevation for cancer was observed only in the first 2 years of follow-up (4.14, 2.74-6.26 versus 1.14, 0.93-1.40 after 2 years), while that for cerebrovascular disease was relatively stable for the entire period. In conclusion, HVE differed according to the causes and length of follow-up.