Pigeons were trained in a delayed matching task in which the samples were short (2 s) and long (10 s) presentations of either a houselight or a keylight. Transfer trials involved short and long presentations of the nontrained signal as the sample. In the intermittent transfer test, infrequent transfer trials were intermixed with more frequent training trials; in the sustained transfer test, all trials were transfer trials. The intermittent test revealed only weak transfer. The sustained test revealed transfer in the first session only in birds which had received pairings of the transfer signal and food prior to testing. However, regardless of whether the transfer signal had been previously paired with food, birds exposed to consistent contingencies between duration and choice across training and testing learned the transfer task more rapidly than birds exposed to inconsistent contingencies. It was concluded that some training in which the transfer signal serves as the sample is required before the durations of a transfer signal are related to the rules relating duration and responding.