Class size is a first-order consideration in the study of education production and education costs. How larger (smaller) classes affect student outcomes is especially relevant to the growth and design of online classes. We study a field experiment in which college students were  randomly assigned to either a large or a small class. All classes were conducted online. Large classes had, on average, about ten percent more students than small classes (mean increase 9.3 percent, standard deviation 4.1 percent). The experiment was conducted at DeVry University, one of the nation's  largest for-profit post-secondary institutions, and included over 100,000 students in nearly 4,000 classes across 102 different undergraduate and graduate courses. First, we examine main effects on student success in the course, and student persistence in college during subsequent terms. Second, theoretical arguments, like Lazear (2001), suggest the effects of class size are likely heterogeneous by student ability. We test whether class size effects are due the increase in number of students per se, or a change in the quality of peers, or both. Finally, we compare the University's direct cost savings from increased class sizes with the implied losses from reduced student persistence.