Abstract Evidence regarding stress as a risk factor for cancer onset is inconsistent. In this study, based on the Japan Public Health Center-based Prospective Study, we enrolled 101,708 participants aged 40–69 years from 1990–1994. The self-reported perceived stress level was collected at baseline and updated through 5-year follow-up. The association between perceived stress and cancer risk was measured by Cox proportional hazards regression model, adjusted for all known confounders. During follow-up (mean = 17.8 years), we identified 17,161 cancer cases. We found no association between baseline perceived stress level and cancer incidence. However, by taking account of the dynamic changes in perceived stress, time-varying analyses revealed a slightly (4–6%) increased overall cancer risk for subjects under elevated perceived stress levels compared to the ‘low stress level’ group. Analyses concerning long-term perceived stress level showed that individuals with constantly high perceived stress level had an 11% (95% confidence interval 1–22%) excess risk for cancer compared to subjects with persistently low stress levels. This association was confined to men (20% excess risk), and was particularly strong among smokers, alcohol drinkers, obese subjects, and subjects without family history of cancer. Therefore, we concluded high perceived stress level might contribute to excess overall cancer incidence among men.