This paper shows that the best known empirical biases of the Black and Scholes (1973) option pricing formula can be explained by investors learning the parameters of the underlying fundamental process. In the context of an equilibrium model where dividend news evolve on a binomial lattice we derive closed-form pricing formulas for European options under Bayesian learning. Learning effects are found to be able to generate asymmetric skews in the implied volatility surface and systematic patterns in the term structure of option prices. We also infer from S\&P 500 index option prices the parameters characterizing the maintained recursive learning process. This allows us to estimate the dynamics of learning and to provide an empirical test for the model.