Attending group therapy sessions is necessary for a group member to have a chance at receiving benefit from the intervention. Group members' perceptions of their group's climate has been linked with important group member outcomes, including session attendance. On the basis of the writings of Curran and Bauer (2011), the current study examined group members' longitudinal ratings of session engagement and decomposed them into between-groups, between-members, and between-sessions components. These components were then used to examine the relationship between group members' ratings of an engaged group climate in the previous session and their attendance the following session. Session attendance in 573 group sessions for 91 group members in 14 Taiwanese counseling groups was modeled in a 3-level hierarchical model (sessions within group members, within groups). Contrary to our hypotheses, between-groups, between-members, and between-sessions components of engagement were not related to session attendance. However, there was a significant interaction between the between-members and between-sessions components of session engagement and group size in predicting session attendance. The likelihood of attendance increased when group members who, on average, rated sessions as being low in engagement uncharacteristically rated a previous session as high in engagement. The likelihood of attendance also increased when group members who, on average, rated sessions as high in engagement uncharacteristically rated the previous session as low in engagement. Larger group sizes amplified these effects. Expectancy (dis)confirmation theory is used to explain these results. Theoretical, research, and clinical implications are discussed.