Stylized evidence suggests that people process information about their own ability in a biased manner. We provide a precise characterization of the nature and extent of these biases. We directly elicit experimental subjects' beliefs about their relative performance on an IQ quiz and track the evolution of these beliefs in response to noisy feedback. Our main result is that subjects update as if they misinterpret the information content of signals, but then process these misinterpreted signals like Bayesians. Specifically, they are asymmetric, over-weighting positive feedback relative to negative, and conservative, updating too little in response to both positive and negative feedback. These biases are substantially less pronounced in a placebo experiment where ego is not at stake, suggesting they are motivated rather than cognitive. Consistent with Bayes' rule, on the other hand, updating is invariant to priors (and over time) and priors are sufficient statistics for past information. Based on these findings, we build a model that theoretically derives the optimal bias of a decision-maker with ego utility and show that it naturally gives rise to both asymmetry and conservatism as complementary strategies in self-confidence management.