We study the impact of online instruction with a field experiment that randomly assigns 1,344 university students to different proportions of online and in-person lectures in multiple introductory courses. Increased online instruction leaves men’s exam performance unaffected but significantly lowers women’s performance, particularly in math-intensive courses. Online instruction also reduces women’s longer-run performance and increases their study dropout. Exploring mechanisms, we find that women exposed to more online lectures report greater difficulty in connecting with peers, less engaging instructors, and lower course satisfaction. Our findings suggest that shifting toward more online instruction may disproportionally harm women.