Problem-based learning (PBL) is a student-centered instructional method, with roots in constructivist theory of learning. Since its origin at McMaster University in Canada, PBL has been implemented in numerous programs across many domains and many educational levels worldwide. In PBL, small groups of 10–12 students learn in the context of meaningful problems that describe observable phenomena or events. The PBL process consists of three phases. The first is the initial discussion phase in which the problem at hand is discussed, based on prior knowledge. This initial phase leads to the formulation of learning issues (i.e., questions) that students will answer during the next phase, the self-study phase. Here, they independently select and study a variety of literature resources. During the third and final phase, the reporting phase, students share their findings with each other and critically evaluate the answers to the learning issues. A tutor guides the first and third phase of the process. PBL is based on principles from cognitive and educational psychology that have demonstrated their capacity to foster learning. More specifically, four principles are incorporated in the PBL process: (a) connection to prior knowledge, (b) collaborative learning among students, but also among teachers, (c) gradual development of autonomy, and (d) a focus on the application and transfer of knowledge. Research on the effects of PBL in terms of knowledge acquisition shows that students in traditional, direct instruction curricula tend to perform better on assessments of basic science knowledge. However, differences between PBL students and students in direct instruction classrooms on knowledge tests tend to diminish over time. There is, however, a lack of controlled experiments in this line of PBL research. Directions for future research should focus on combining the best of both direct and student-centered instruction, explore the possibilities of hybrid forms, and investigate how the alignment of scale and didactics of an instructional method could be optimized.