Cancer morbidity and mortality vary among socio-economic groups. We investigated whether educational differences in behaviour could explain the variation in cancer incidence. In 1991, a postal questionnaire on socio-economic status, exposure variables, health and health-related behaviour was filled out by 18,973 participants (response rate 70.1%) of the longitudinal GLOBE study. Participants were followed and linked with the regional population-based Eindhoven Cancer Registry. Between 1991 and 1998 a total of 760 new tumours were found. The risk of cancer (all sites combined) was higher in the three quartiles of lower educational level compared with the highest educational level, odds ratios (ORs) varying from 1.25, 1.34, 1.27 to 1.00 from the lowest to the highest category, respectively (P=0.14). The relative risk (RR) for lung cancer for low versus high education was 2.7 [95% confidence interval (CI) 1.3-5.3]; adjustment for smoking, alcohol intake and physical activity decreased the risk to 1.6 (95% CI 0.8-3.3). Smoking alone explained 39% of the association, when alcohol intake and physical exercise were added to the model 61% of the effect was explained. In conclusion, a lower education is associated with increased cancer risks, which can be explained partly by behavioural factors.