We conducted a cohort study for 2 years to examine the causal relationship between perceived job stress and mental health. Questionnaire surveys, including a 30-item General Health Questionnaire (GHQ) and a questionnaire on perceived job stress were carried out every 6 months for 2 years. To clarify the causal relationship between job stress and mental health, we followed a group of workers who initially had a GHQ score < or = 7. Out of 462 workers who were thought to be in a healthy mental state, 282 were successfully followed for 2 years. We considered subjects who developed unhealthy mental health states (GHQ score > or = 8) as hazardous cases. To control potential confounding factors, proportional hazard analysis was done. The overall proportion hazardous cases detected in the development of an unhealthy mental health state over two years was 55.7%. Using Cox's proportional hazard model, workers who complained of perceived job stress had a greater hazard than those without job stress. In particular, the item 'poor relationship with superior' showed the largest adjusted hazard ratio [95% confidence interval (CI)] of 1.51 (1.06-2.15). The item 'too much trouble at work' also had a significant hazardous effect on mental health with an adjusted hazard ratio (95% CI) of 1.43 (1.00-2.04). Some specific items of perceived job stress could cause mental ill health in workers.