The ability to infer causal structure from data is central to the growth of both scientific and everyday knowledge. Traditional explanations of human causal induction have tended to emphasize either domain-general covariation-based learning, or domain-specific knowledge about causal mechanisms. However, it is clear that both of these factors interact in most interesting cases of causal induction - the key questions are what prior knowledge is used, and how it interacts with statistical inference.
I will present a computational framework that addresses these two questions, formulating the problem of causal induction as a Bayesian decision among a set of causal models generated by a domain theory. I will apply this framework to two phenomena of causal induction - inferring causal relationships from contingency data and learning the structure of physical systems.