Background: Subjective survival probabilities (SSPs) are a good predictor of mortality, go beyond the aggregate description of survival defined by life tables, and are important for individuals’ decision-making in later life. Despite the well-known mortality differentials by education as well as by characteristics such as smoking, little investigation has focused on SSPs by population sub-groups. Methods: We use data on individuals aged 50-89 from the Health and Retirement Study(HRS) carried out in the USA between 2000 and 2012 (N=23,895). Each respondent was asked to assess the probability to survive to a given target age according to their age at the time of the survey. We assess how individuals’ SSPs and estimated objective survival probabilities (OSPs) vary by education and smoking and calculate, for each respondent, the gap between them. Results: Consistently with real mortality patterns, smokers report the lowest SSPs, both among lower and higher educated people. When comparing SSPs and OSPs we find that, irrespectively of the smoking status, higher educated people are more likely to correctly predict their survival probabilities than their lower educated counterparts. Within both education groups, past smokers better predict their survival probability. Current smokers with low education show the highest probability to overestimate their survival probability. Conclusions: Lower educated people and smokers are aware of their lower life expectancy. Still, they overestimate their survival probabilities more than the higher educated and non-smokers. Our findings emphasize the need for policy makers to disseminate information about the risks of smoking, targeting people with lower education.