While self-assessment is a widely explored area in educational research, our understanding of how students assess themselves, or in other words, generate self-feedback, is quite limited. Self-assessment process has been a black box that recent research is trying to open. This study explored and integrated two data collections (secondary and higher education) that investigated students' real actions while self-assessing, aiming to disentangle self-assessment into more precise actions. Our goal was to identify self-assessment processes and profiles to better understand what happens when students self-assess and to design and implement better interventions. By combining such data, we were able to explore the differences between secondary and higher education students, the effects of external feedback on self-assessment, and to propose a model of ideal self-assessment (SEFEMO). Using think-aloud protocols, direct observation and self-reported data, we identified six main actions (read, recall, compare, rate, assess, and redo) and four self-assessment profiles. In general, secondary and higher education students showed the same actions and very similar profiles. External feedback had a negative effect on the self-assessment actions except for the less advanced self-assessors. Based on data from more than 500 self-assessment performances, we propose a model of self-feedback.