Abstract: Causal learning is the mechanism by which an intelligent system discovers how the world works. The psychological debate about human causal learning has focused on the distinction between covariation and causation and has been carried out in the context of two probabilistic models that both yield estimates of causal strength: The ΔP rule (Jenkins & Ward, 1965), which strictly assesses covariation, and the power-PC theory (Cheng, 1997), which specifies causal assumptions that serve as a framework for interpreting contingency information. Previous research has shown evidence for both approaches; i.e., causal judgments vary with causal power when ΔP is constant and with ΔP when causal power is held constant. According to Tenenbaum and Griffiths (2001), both ΔP and power can be seen as maximum likelihood parameter estimates on a fixed graph, in which a link exists between the cause and the effect. In their 'causal support' model, such estimates are preceded by a Bayesian inference about whether the causal link exists. Tenenbaum and Griffiths argue that failure to incorporate uncertainty is responsible for the inability of previous psychological models to fully capture causal judgments. In contrast, causal support reflects a combination of the parameter estimate and the degree of certainty in that estimate. Given a power parameterization, the causal support model predicts, among other things, the perplexing interaction between power and ΔP. In this thesis, I sought to elucidate the relationship between structure inference, strength estimation and causal assumptions. I found that the ability of the causal support model to account for structure inference depends critically on its adoption of the causal power assumptions (i.e., on parameterization). Moreover, clear dissociations were demonstrated between strength estimates and certainty in (1) estimates and (2) in the existence of a link. These dissociations provided empirical evidence for the notion that variations in certainty can explain why judgments sometimes deviate from causal power in the direction predicted by ΔP. Finally, inferences about structure in complex environments were found to be based on causal power, rather than causal support, or ΔP. The implications of these results for a 'distal stimulus' interpretation of causal power are discussed.