Recognition memory for arbitrary complex stimuli was measured using the operating characteristic technique developed by Egan [J. P. Egan, AFCRC TN 58-51, AD 152650 (1958)]. Ten complex stimuli (500 ms in duration), each consisting of six, randomly chosen, equal amplitude sinusoidal components spanning frequencies from 300 to 3000 Hz, were used as a training set. After 15 min of listening to the training set, listeners were presented a 20-stimulus test set consisting of the ten training stimuli along with ten additional complex stimuli (generated in the same way as the ten training stimuli). In a recognition memory paradigm, listeners used a five-point rating scale to rate their confidence that a stimulus presented from the test set was from the training set. Different constraints were placed on the selection of the complex waveforms to determine how those constraints affected recognition memory. For one condition there were no constraints except as described above; for a second condition the six-tonal components were harmonics (randomly chosen) of a common fundamental; and for a third condition the six-tone complex was sinusoidally amplitude modulated at rates of 4, 16, or 32 Hz. The results will be discussed in terms of auditory object perception and the use of receiver operating techniques to measure recognition memory. [Work supported by a Program Project Grant from the NINCDS and a grant from AFOSR.]