Categories of self-reported health (SRH) are often used as a measure of health status. However, the difficulties with measuring health mean that getting individuals to select into SRH categories may involve misclassification. In the current study, we explore this issue using the Household, Income and Labour Dynamics in Australia (HILDA) Survey where a panel of individuals are asked the same question about their health status twice in some years. We find that a considerable number of individuals switch their stated health category when asked a second time each year. We first analyse whether this switching is related to any particular individual characteristics and then explore possible approaches to correct for any bias introduced from the misclassification when SRH is used to explain the chance of survival. We also use these approaches to estimate and predict survival for our population. We find that SRH is likely measured with considerable error and this is likely to have consequences for using it as an explanatory variable in models. Having multiple questions on health status in a survey or using answers from multiple years in a longitudinal panel survey may improve the precision of the estimated effect and the prediction of SRH on survival for individuals.