In this paper, we analyze class size effects on college students exploiting data from a project offering special remedial courses in mathematics and language skills to freshmen enrolled at an Italian medium-sized public university. To estimate the effects of class size, we exploit the fact that students and teachers are virtually randomly assigned to teaching classes of different sizes. From our analysis, it emerges that controlling for a number of individual characteristics, larger classes determine a significant and sizeable negative effect on student performance in mathematics. Importantly, this negative effect is significantly larger for low-ability students and negligible for high-ability ones. On the other hand, class size effects do not appear to be relevant for student achievement in language skills. (Contains 2 figures, 5 tables, and 3 notes.)