Research indicates many professional development (PD) programs fail to promote a significant change in science teaching, despite the resources invested in them. While previous research addresses this problem through aspects of policy making, design, and teachers' identity, we suggest a focus on teachers' engagement in learning during PD sessions. Engagement is critical for learning processes and outcomes and affects teachers' professional learning, and, hence, their teaching. Yet little research explores teachers' engagement during professional learning. Most researchers approach engagement from a psychological perspective, using self‐report tools, usually questionnaires. To enable a more complete understanding of science teacher engagement in PD, we offer a novel multimodal contextual perspective on engagement, one that perceives engagement as embodied, situated, local, and dynamic. We present a multimodal situated analytical tool referring to the expected behavior in specific contexts, as defined by the norms of the situation and its participants. To develop the tool and apply it, we used video‐recorded PD sessions of Israeli elementary and middle school science teachers, focusing our analysis on the engagement of 10 teachers over 21.5 h. We used multimodal analysis to map teachers' engagement in relation to the requested action, referring to: (1) five engagement levels determined by the expected behavior; (2) different engagement modalities—verbal, body posture and gestures, gaze, and facial expressions; and (3) engagement duration. We demonstrate the potential of this analytical framework to advance the understanding of PD interactions by comparing: (1) two teachers' engagement in the same activity; and (2) teachers' engagement in three PD activities: whole‐class learning, game playing, and inquiry‐into‐practice. The analysis reveals differences within and between teachers, as well as between the activities, in levels of engagement and alignment between modalities. We discuss the theoretical and practical implications of these findings and suggest how the tool may be applied in research.