Self-assessment has been portrayed as a way to promote lifelong learning in higher education. While most of the previous literature builds on the idea of self-assessment as a formative tool for learning, some scholars have suggested using it in a summative way. In the present study, we have empirically compared formative and summative models for self-assessment, based on different educational purposes (N = 299). Latent profile analysis was used to observe student subgroups in terms of deep and surface approaches to learning. The results show that the student profiles varied between the self-assessment models. The students taking part in the summative self-assessment group were overrepresented amongst the profile with high level of deep approach to learning. Also, summative self-assessment was related to an increased level of self-efficacy. The study implies that summative self-assessment can be used to foster students' studying; however, this requires a context where aligning self-assessment with future-driven pedagogical purposes is possible.