This paper models an agent in a multi-period setting who does not update according to Bayes. Rule, and who is self-aware and anticipates her updating behavior when formulating plans. Choice-theoretic axiomatic foundations are provided. Then the model is specialized axiomatically to capture updating biases that re.ect excessive weight given to (i) prior be- liefs, or alternatively, (ii) the realized sample. Finally, the paper describes a counterpart of the exchangeable Bayesian model, where the agent tries to learn about parameters, and some answers are provided to the question, "what does a non-Bayesian updater learn?"