Although venture teams whose founders are dissimilar (heterophilious) tend to outperform teams whose founders are similar (homophilious), most new venture teams are characterized by homophily. I try to explain this puzzle with a learning model in which founders are prone to two cognitive biases: overoptimism and self-serving attributions. Founders choose cofounders with similar beliefs as themselves because they expect this to promote the most effective allocation of effort to the venture. Self-serving bias reinforces and perpetuates these beliefs. In principle, informed outsiders (e.g., practitioners or hands on investors) can improve venture team composition compared with private choices by founders.