Recent research has demonstrated that retirement planning and well-being are closely tied to probabilistic forecasts about future events. Using longitudinal data from the Health and Retirement Study, I show that individuals’ subjective survival forecasts exhibit systematic biases relative to life table data. In particular, many respondents fail to account for increases in yearly mortality rates with age, both longitudinally and in crosssection. Additionally, successive cohorts of the near elderly do not appear to revise survival forecasts to match increases in longevity. Forecasting bias may merely be due to the framing of questions designed to elicit expectations, but real biases may result in suboptimal savings rates and timing of retirement. Cross-sectional variation in subjective survival forecasts also appears to reflect differences in cognitive ability across respondents, suggesting that subjective information is more relevant for some individuals than others. Despite these shortcomings, subjective mortality probabilities predict actual mortality and portfolio choice, and they contain information not found in selfreported health status or objective measures of health limitations.