Evidence-based teaching practices (EBTP)—like inquiry-based learning, inclusive teaching, and active learning—have been shown to benefit all students, especially women, first-generation, and traditionally minoritized students in science fields. However, little research has focused on how best to train teaching assistants (TAs) to use EBTP or on which components of professional development are most important. We designed and experimentally manipulated a series of pre-semester workshops on active learning (AL), dividing subjects into two groups. The Activity group worked in teams to learn an AL technique with a workshop facilitator. These teams then modeled the activity with their peers acting as students. In the Evidence group, facilitators modeled the activities with all TAs acting as students. We used a mixed-methods research design (specifically, concurrent triangulation) to interpret pre- and post-workshop and post-semester survey responses. We found that Evidence group participants reported greater knowledge of AL after the workshop than Activity group participants. Activity group participants, on the other hand, found all of the AL techniques more useful than Evidence group participants. These results suggest that actually modeling AL techniques made them more useful to TAs than simply experiencing the same techniques as students—even with the accompanying evidence. This outcome has broad implications for how we provide professional development sessions to TAs and potentially to faculty.