Background: There has been an ongoing debate about the inconsistent effects of heterogeneous ability grouping on students in small group work such as project-based learning. Aim: The present research investigated the roles of group heterogeneity and processes in project-based learning. At the student level, we examined the interaction effect between students' within-group achievement and group processes on their self- and collective efficacy. At the group level, we examined how group heterogeneity was associated with the average self- and collective efficacy reported by the groups. Sample: The participants were 1,921 Hong Kong secondary students in 367 project-based learning groups. Method: Student achievement was determined by school examination marks. Group processes, self-efficacy and collective efficacy were measured by a student-report questionnaire. Hierarchical linear modelling was used to analyse the nested data. Results: When individual students in each group were taken as the unit of analysis, results indicated an interaction effect of group processes and students' within-group achievement on the discrepancy between collective- and self-efficacy. When compared with low achievers, high achievers reported lower collective efficacy than self-efficacy when group processes were of low quality. However, both low and high achievers reported higher collective efficacy than self-efficacy when group processes were of high quality. With 367 groups taken as the unit of analysis, the results showed that group heterogeneity, group gender composition and group size were not related to the discrepancy between collective- and self-efficacy reported by the students. Conclusions: Group heterogeneity was not a determinant factor in students' learning efficacy. Instead, the quality of group processes played a pivotal role because both high and low achievers were able to benefit when group processes were of high quality.