The main goal of this study was to illustrate and provide some direction for dealing with the complexities of propensity score matching within different multilevel contexts. Special attention is given to how procedures typically applied in a non-hierarchical setting may be modified to properly reduce the expected bias in the estimated treatment effect of a high school-level intervention on college-level outcomes. In particular, students self-selected into a high school level intervention and the outcome of interest was observed within the context of the college or university in which students subsequently enrolled. A simulation was prepared to delineate the relevant issues and to demonstrate the effects of ignoring a hierarchical nesting of subjects. An empirical example of an evaluation of a large-scale, national educational program is given.