Members of a health maintenance organization (N = 353) interacted with a computer program that provided personalized information about their risk of developing colon cancer in the next 20 years. Prior to computer feedback, most people greatly overestimated their numerical, absolute risk (chances per 1000) and also overestimated their relative risk compared to peers (e.g., “above average”). Their relative risk estimates were correlated with several risk factors, whereas their absolute risk estimates were not, suggesting that assessing individual risk perceptions with numerical, absolute risk scales may provide misleading information about what people believe. Computer feedback improved the accuracy of mean risk estimates, but about half of participants did not accept the personalized feedback as correct. In fact, correlations between actual and perceived risk were no greater among participants who received risk scores than among those who did not. Three possible explanations for resistance to lower-than-expected risk feedback are considered.