This study examined fidelity of implementation in a randomized trial of Banking Time, a classroom-based intervention intended to improve children's behavior, specifically for those at risk for developing externalizing behavior problems, through improving the quality of teacher-child interactions. The study sample comes from a randomized controlled trial that took place over three years and included 183 teachers and 470 preschool children recruited across three sites. Results exploring fidelity of implementation indicated that Banking Time practices were present across treatment and nontreatment groups. In addition, the presence of more Positive Banking Time Practices accounted for improvements in teacher-child interactions while Restricted Banking Time Practices showed inconsistent patterns, relating both to positive and negative teacher-child interactions. Further, random assignment to Banking Time resulted in the presence of more Positive Banking Time Practices and fewer Restricted Banking Time Practices, which accounted for the improved teacher-child interactions. Thus, findings point to the critical need to better account for intervention fidelity across conditions in randomized trials, allowing for better understanding of the mechanism through which interventions lead to intended outcomes. (author abstract)