Objective Using a novel mediation method that presents unbiased results even in the presence of exposure-mediator interactions, this study estimated the extent to which working conditions and health behaviors contribute to educational inequalities in self-rated health in the workforce. Methods Respondents of the longitudinal Survey of Health, Ageing, and Retirement in Europe (SHARE) in 16 countries were selected, aged 50-64 years, in paid employment at baseline and with information on education and self-rated health (N=15 028). Education, health behaviors [including body mass index (BMI)] and working conditions were measured at baseline and self-rated health at baseline and two-year follow-up. Causal mediation analysis with inverse odds weighting was used to estimate the total effect of education on self-rated health, decomposed into a natural direct effect (NDE) and natural indirect effect (NIE). Results Lower educated workers were more likely to perceive their health as poor than higher educated workers [relative risk (RR) 1.48, 95% confidence interval (CI) 1.37-1.60]. They were also more likely to have unfavorable working conditions and unhealthy behaviors, except for alcohol consumption. When all working conditions were included, the remaining NDE was RR 1.30 (95% CI 1.15-1.44). When BMI and health behaviors were included, the remaining NDE was RR 1.40 (95% CI 1.27-1.54). Working conditions explained 38% and health behaviors and BMI explained 16% of educational inequalities in health. Including all mediators explained 64% of educational inequalities in self-rated health. Conclusions Working conditions and health behaviors explain over half of the educational inequalities in self-rated health. To reduce health inequalities, improving working conditions seems to be more important than introducing health promotion programs in the workforce.