This study examines the development of a specific sub-skill for studying and improving teaching--the generation of hypotheses about the effects of teaching on student learning. Two groups of elementary preservice teachers (PSTs) were compared: one group that attended a typical mathematics-methods course and one that attended a course integrating analysis skills for learning from teaching. Data consist of PSTs' comments on video clips of mathematics instruction administered before and after course completion. Findings reveal that PSTs at the beginning of the program struggled to generate hypotheses with relevant evidence, often equating teacher behavior or student correct answers as evidence of student understanding. After course participation, PSTs who attended the course with integrated analysis skills significantly improved in their ability to generate hypotheses based on student evidence whereas their counterparts continued to display difficulties. Implications for teacher education and future research are considered.