Scholars have long argued that principals should be "instructional leaders," but few studies have empirically linked specific instructional leadership behaviors to school performance. This study examines the associations between leadership behaviors and student achievement gains using a unique data source: in-person, full-day observations of approximately 100 urban principals collected over 3 school years. We find that principals' time spent broadly on instructional functions does not predict student achievement growth. Aggregating across leadership behaviors, however, masks that some specific instructional investments predict year-to-year gains. In particular, time spent on teacher coaching, evaluation, and developing the school's educational program predict positive achievement gains. In contrast, time spent on informal classroom walkthroughs negatively predicts student growth, particularly in high schools. Additional survey and interview evidence suggests this negative association may arise because principals often do not use walkthroughs as part of a broader school improvement strategy.