Educational institutions face pressing challenges regarding student persistence, time to graduation, and underrepresentation of women and minorities in STEM fields (1–5). Developing targeted and effective solutions to these problems requires a concrete understanding of how diverse student groups progress through academic programs. In light of this, there are growing calls for a new science of educational pathways (6), but this idea remains more metaphor than methodology. Transcript data hold the promise of revealing the paths students take through a curriculum (7, 8), but existing frameworks do not provide a fine-grained, processual account of how students arrive at their academic destinations. In this study, we present a theoretically grounded, data-driven framework for translating transcript data into academic pathways. Our framework delivers information about students’ movements both through the space of possible majors and within a particular program. This information is remarkably detailed, but its richness creates statistical challenges in that the analyst must allow for temporal dynamics, diverse pathways, and the possibility that the most likely path for a given type of student differs across contexts (e.g., fields of study, colleges, or universities). We develop a question- and data-driven statistical model that leverages the richness of pathways data, with each level tuned to nonparametrically extract a different kind of information about trajectories, student demographics, and how their relationship varies across contexts. We apply this framework to data from a large public university to reveal how students of varying backgrounds, including historically underrepresented groups, enter and exit fields of study.