We develop a theory of “Partial Equilibrium Thinking” (PET), a type of misinference whereby agents fail to understand the general equilibrium consequences of their actions when inferring information from endogenous outcomes. PET generates a two-way feedback between outcomes and beliefs, which can lead to arbitrarily large deviations from fundamentals. In financial markets, PET equilibrium outcomes exhibit over-reaction, excess volatility, high trading volume, and return predictability. We draw a distinction between models of misinference and models with biases in Bayesian updating, and study how these two departures from rationality interact. We show that misinference from mistakenly assuming the world is rational can vastly amplify biases in Bayesian updating, and that the distinction between these two biases can have important quantitative implications.