Expectations about health and future needs for healthcare potentially influence decisions about retirement, savings, long-term care insurance, and much else. Inaccurate health expectations can result in suboptimal decisions, loss of welfare and potentially contribute to inequalities in health and related outcomes. In this paper, we will assess and decompose the accuracy of four expectations about health-related events – working full-time after the age of 62 and 65, onset of a work-limiting health problem within 10 years and admission to a nursing home within 5 years – and investigate its heterogeneity by education. We use the mean squared prediction error as a measure of prediction accuracy, and the Yates decomposition of it into bias, discrimination, outcome uncertainty and noise. The results indicate that individuals are on average highly inaccurate in their predictions. For all events, subjective expectations reported by the lower educated are less accurate. Explanations vary somewhat, but the predictions of the higher educated tend to contain less noise and greater discriminatory power, while their outcome variance is often lower. Overall, these results indicate that individuals, and especially the lower educated, might not be planning optimally. Additionally, the education gradient in prediction accuracy points to a mechanism through which inequalities in health and wellbeing may be exacerbated.