The present experiment examined whether discrimination learning shapes the single-peaked response distributions usually obtained with the peak procedure. Two sources of learning in pigeons were disclosed: learning to respond near the time of reinforcement on fixed interval (FI) trials and learning to withhold responding once the FI duration had elapsed on peak interval (PI) trials. Pigeons also produced a highly unexpected second peak in responding on nonreinforced PI trials at 3 times the FI duration. Follow-up experiments showed that a 1:4 FI:PI duration ratio supported double peaks, but only 1 peak was obtained with a 1:8 FI:PI duration ratio. Finally, 4 peaks could be observed on extra-long PI trials under a 1:4:8 FI:PI:PI ratio procedure. The multiple-peaked response distributions are an unprecedented finding that present a major challenge to any theory of time perception.