Through the use of event-history techniques, we will show that a duration framework is adapted to the analysis of higher education attrition. Our dropout model allows for estimates to vary over time. While some factors exhibit constant effects, like high school characteristics, other effects do vary from the first year to the fourth. Men and women do not generally exhibit the same dropout behaviour. Socio-economic background, and especially the parents' level of education, would appear to be influential only at the beginning of the university period. Results suggest that only well-targeted actions considering both the "early leavers" and "late leavers" could serve to increase the internal efficiency of French universities. (Contains 1 note, 3 tables, and 1 figure.)