This paper explores model misspecification in an observational learning framework. Individuals learn from diverse sources of information, including private and public signals and the actions of others. However, they may not know the true model that generates these signals, or how other individuals' actions reflect their private information. An agent's type specifies her model of the world; misspecified types have incorrect beliefs about the signal distribution and how other agents draw inference. First, we establish that the correctly specified model is robust in that agents with approximately correct models asymptotically learn the true state. Second, we develop a simple criterion to identify what asymptotic learning outcomes arise when misspecification is more severe. Depending on the nature of the misspecification, learning may be correct, incorrect or beliefs may not converge. Different types may asymptotically disagree, despite observing the same information. This framework captures behavioral biases such as confirmation bias, underweighting or overweighting information, partisan bias and correlation neglect, as well as models of inference such as level-k and cognitive hierarchy.