Abstract We extend Pearl's criticisms of principal stratification analysis as a method for interpreting and adjusting for intermediate variables in a causal analysis. We argue that this can be meaningful only in those rare cases that involve strong functional dependence, and even then may not be appropriate.