This study aims to assess empirically whether being employed or returning to work is beneficial for all in terms of mental health, especially for those who already suffer from a longstanding illness or disability. We use longitudinal surveys from Australia, Canada, Switzerland and the UK to estimate panel data models that link decisions regarding labour market choices to health developments. To allow for state dependence of mental health, a dynamic panel model is used. The longitudinal analysis shows that non-employment generally is worse for mental health than working. The mental-health payoff to employment varies depending on the type of employment contract and working conditions. In particular, the mental health benefits for inactive individuals who obtain a non-standard job appear to be smaller than for those moving into standard employment arrangements, even after controlling for pre-existing mental health problems.