A bottlenose dolphin, already highly proficient in two‐choice auditory discriminations, was trained over a nine‐day period on auditory delayed matching‐to‐sample and then tested on 346 unique matching problems, as a function of the delay between the sample and test sounds. Each problem used new sounds and was from five to 10 trials long, with the same sound used as the sample for all trials of a problem. At each trial, the sample was projected underwater for 2.5 sec, followed by a delay and then by a sequence of two 2.5‐sec duration test sounds. One of the test sounds matched the sample and was randomly first or second in the sequence, and randomly appeared at either a left or right speaker. Responses to the locus of the matching test sound were reinforced. Over nine, varying‐sized blocks of problems, the longest delay of a set of delays in a block was progressively increased from 15 sec initially to a final value of 120 sec. There was a progressive increase across the early blocks in the percentage of correct Trial 1 responses. A ceiling‐level of 100% correct responses was then attained over the final six blocks, during which there were 169 successive Trial 1 responses bracketed by two Trial 1 errors (at 24‐ and 120‐sec delays). Performance on trials beyond the first followed a similar trend. Finally, when the sample duration was decreased to 0.2 sec or less, matching performance on Trial 1 of new problems dropped