This paper examines how learning and competition shape brokerage commission fees, using a plain-vanilla setting from China’s brokerage industry. We document three key facts: substantial price dispersion across investors, naïve investors paying higher fees, and widening dispersion over time. To explain these patterns, we develop a dynamic model where brokers compete for naïve and sophisticated investors while learning their types through repeated interactions. The model implies that more broker competition can backfire, raising markups. Using branch-level data, we provide causal evidence supporting this prediction. Counterfactuals show that, in markets with heterogeneous investors, improving financial literacy compresses markups more effectively than broker entry, highlighting the importance of investor education.