With the increased number of schools adopting social-emotional learning (SEL) programming, there is increased emphasis on the role of implementation in obtaining desired outcomes. Despite this, the current knowledge of the active ingredients of SEL programming is lacking, and there is a need to move from a focus on ‘whether’ implementation matters to ‘what’ aspects of implementation matter. To address this gap, the current study utilizes a latent class approach with data from year 1 of a randomized controlled trial of Second Step (61 schools, 321 teachers, over 7300 students). Latent classes of implementation were identified, then used to predict student outcomes. Teachers reported on multiple dimensions of implementation (adherence, dosage, competency), as well as student outcomes. Observational data were also used to assess classroom behavior (academic engagement and disruptive behavior). Results suggest that a three-class model fits the data best, labeled as high-quality, low-engagement, and low-adherence classes. Only the low-engagement class showed significant associations with poorer outcomes, when compared to the high-quality class (not the low-adherence class). Findings are discussed in terms of implications for program development and implementation science more broadly.