Recent work on spectral profile analysis demonstrates that observers can extract amplitude relationships between components in complex stimuli. The experiments to be discussed demonstrate that observers also can extract frequency relationships, specifically, they can discriminate frequency ratios of components in a two-tone complex. The observer is presented with two complexes, one whose components are at frequencies of f and fr, the other with components at f and f(r + Δt), and, in a 2IFC task, is required to choose the observation interval containing the larger ratio. The base frequency f is random both within and across trials; thus, the observer must, in effect, compare the frequencies within each complex to perform well. Good performance is shown with small values of Δr: for example, with r = 1.25 a Δr of 0.017 yields approximately 71% correct when f is randomized over a 400–800 Hz range. This is comparable to the performance shown by trained musicians in discrimination of musical intervals when each of the two tones comprising the interval are presented successively [E. M. Burns and W. D. Ward, J. Acoust. Soc. Am. 63, 456–468 (1978)]. The paper will summarize our initial findings on the effects of varying r and frequency range, and on the form of the psychometric function. [Supported by NINCDS grants NS12125 and NS07889.]