Background: The predominant dissertation model used in teacher education courses in Flanders is the "literature study with practical processing". Despite the practical supplement, this traditional model does not fit sufficiently well with autonomous learning as the objective of modern teacher education dissertations. This study reports on the development of a portfolio model to serve as an alternative to the current model and afford greater opportunities for autonomous learning. Purpose: This quasi-experimental study aimed to examine the usefulness of portfolio as a means of promoting autonomous learning. Sample: Two teacher education courses collaborated, involving 174 teacher education students and 44 supervisors in this study. Design and methods: Using a pre- and post-test quasi-experimental design, we examined whether the use of portfolio as a dissertation model has a greater effect on the students' capacity for autonomous learning than the literature study with practical processing, using three existing questionnaires. Differences between the experimental group and the control group were revealed by means of covariance analysis. Results: Our research shows that the portfolio model results in students acquiring greater metacognitive knowledge. It also appears that students only get the chance to use their metacognitive skills when supervisors give them sufficient autonomy. However, supervisors experience loss of control and therefore tend to restrict students' autonomy. Conclusions: This leads us to the paradox of autonomous learning: students will only learn autonomously when they are given the freedom to learn autonomously. Supervisors need to transfer learner control. (Contains 16 tables.)