We study a sequential model of Bayesian social learning in networks in which agents have heterogeneous preferences, and neighbors tend to have similar preferences — a phenomenon known as homophily. We find that the density of network connections determines the impact of preference diversity...