This research uses the Household, Income and Labour Dynamics in Australia (HILDA) Survey to investigate the impacts of health on labour force status of older working-age Australian men. We estimate a model that exploits the longitudinal nature of the data and takes the correlation between the two error terms in the health and labour force status equations into account. The results show that controlling for unobserved heterogeneity and the correlation between the two equations is important. It is also found that any restriction on the correlation between the two equations appears to lead to underestimation of the direct health effects.