We compare the explanatory power of four payoff-based learning models on two data sets where agents have extremely limited information. Under the serial mechanism in Chen (1998), the payoff-assessment learning model performs significantly better than the other three models, followed by EWA, which in turn is followed by reinforcement learning and the responsive learning automata. Under the average cost pricing mechanism, however, the performance of the four models are statistically indistinguishable. They are all close to the performance of the random choice model.