California Psychotherapy Alliance Scale–Group (CALPAS-G) data were collected from 1,138 group sessions attended by 248 group members in 16 counseling groups. Hierarchical linear modeling (HLM) was used to derive between-groups, between-member, and between-session variance components and intraclass correlation coefficients (ICCs) for the 12 CALPAS-G items. Using Ledermann and Kenny's (2012) descriptions of variable types, we examined differences in between-groups variance for the 6 CALPAS-G items classified as “Individual” items and the 6 CALPAS-G items classified as “Group” items. A Related-Samples Wilcoxon's Signed Ranked Test showed that the ICCs for the Group items were significantly larger than the ICCs for the Individual items. The results show the importance of how items are worded. If researchers want to accurately examine the between-groups component of the group therapy relationship they should develop measures that ask clients to describe their perceptions of the group, not the members own experience of the group.