This study employed a multisite design to investigate the differential impact of deductive and guided inductive instruction for second language (L2) grammar development in ecologically valid classroom contexts. Students (n = 138) from eight intact third-year L2 Spanish classes in three public high schools in the United States received deductive instruction (n = 49), guided inductive instruction via PACE (n = 49), or no explicit instruction (n = 40) on the pronoun "se" in non-agentive constructions. Learning was measured with pre-, post-, and delayed posttests of auditory acceptability judgment and written production tasks. Multilevel models indicated that both instructional groups evidenced L2 development in producing "se" in non-agentive contexts, but only deductive groups were more accurate than controls for acceptability judgments. Findings also provided evidence of overgeneralization of "se," especially for deductive groups. Intraclass correlation coefficients revealed variance related to the socio-local contexts of classrooms, teachers, and schools.