Four groups of pigeons were trained to perform a delayed matching‐to‐sample task with a single delay of 0, 2, 4, or 6 s from the outset of training. The longer the training delay, the more sessions were required for all birds to reach the same level of response accuracy. Following initial training, five test sessions that included nonreinforced trials with delay intervals of 0, 2, 4, 6, 8, and 10 s were interspersed between training sessions. Unlike typical forgetting functions in which accuracy decreases monotonically with increasing delay, the forgetting functions from test sessions resembled generalization gradients with the peak of the functions occurring at the training delay. Following additional training for all birds with a 0‐s delay, forgetting functions decreased monotonically with increasing delay. The results suggested that remembering can be trained at a specific delay interval, and generalizes to similar delay intervals. Generalization along the temporal dimension of delay may contribute to typical forgetting functions in which accuracy decreases from 0‐s delay.