We prove an N2-o(1) lower bound on the randomized communication complexity of finding an epsilon-approximate Nash equilibrium (for constant epsilon > 0) in a two-player N x N game.