Objectives. To determine whether perceived job stress affects mental health in occupational settings. Methods. A 2 year cohort study was conducted. Initially, a survey including the general health questionnaire (GHQ) and a questionnaire about perceived job stress was carried out. Of 462 workers who initially showed a GHQ score of ≤ 7 310 were successfully followed up for 2 years. The 2 year risks of developing mental ill health (a GHQ score ≥ 8) were assessed relative to perceived job stress. To control for potential confounding factors, multiple logistic regression analyses were conducted. Results. The overall 2 year risk for developing mental ill health was high at 57.7%. Workers who reported aspects of perceived job stress showed a greater 2 year risk than those without stress. Multiple logistic regression analyses showed that some components of perceived job stress were associated with a higher 2 year risk, among which 'not allowed to make mistakes' showed the largest adjusted odds ratio (OR) (95% confidence interval (95% CI)) of 2.37 (1.32 to 4.29). 'Poor relationship with superior' had a significant effect on mental health only in women, with an adjusted OR (95% CI) of 3.79 (1.65 to 8.73). Conclusions. Certain specific items of perceived job stress seem to be associated with mental ill health in workers. These could broadly be described as job strain, or job demand items. The type of job stress that predicts mental health may be dependent on the characteristics of the workplace investigated.