The credibility revolution has promoted the adoption of research designs that permit identification and estimation of causal effects. Understanding which mechanisms drive measured causal effects remains a challenge. A dominant current approach to the quantitative evaluation of mechanisms relies on the detection of heterogeneous treatment effects with respect to pre-treatment covariates. This paper develops a framework to understand when such heterogeneous treatment effects can support substantive inferences about the activation of a mechanism. We show first that this design does not provide evidence of mechanism activation without additional assumptions. Further, even when these assumptions are satisfied, if a measured outcome is produced by a non-linear transformation of a latent variable of theoretical interest, heterogeneous treatment effects are not necessarily informative of mechanisms. We provide new guidance for interpretation and research design in light of these findings.