Pigeons were presented with an analog of a radial-arm-maze task involving 5 response keys. In Experiment 1, pigeons were exposed to either a linear (L) or a two-dimensional matrix (M) array of the 5 keys, and the keys were either each of a distinctive hue (H) or were all white (W). Acquisition was facilitated both by M and by H. In Experiment 2, increasing the number of pecks that constituted a choice from 5 to 20 for half of the LW birds had little effect on performance. In Experiment 3, a delay was interpolated after a number of choices that varied across trials. All groups except the LW 5-peck group showed little disruption in performance with increasing delay (up to a maximum of 1 hr). Inverted-U-shaped point-of-delay interpolation functions were found for each group. The data suggest that pigeons retrospectively coded choices already made when the delay occurred early in the trial and prospectively coded choices yet to be made when the delay occurred late in the trial.