We study the impact of network architecture on the efficiency of information transmission and the dynamics of learning in networks. In our setting, people are positioned into a fixed network and know its structure. All network members receive a partially informative signal about the unknown state and can revise their state estimate based on neighbors’ opinions. Participants have a common interest of learning the state but observe different information based on their network position. We design a novel experimental interface that allows studying the interplay between network architecture and information diffusion in large networks in a controlled laboratory environment. We ask: How do network structures affect the likelihood of reaching consensus? Conditional on reaching the consensus, how likely agents are to choose the correct action? How fast does convergence occur? Is it possible to observe connected networks in which agents agree to disagree?