A relation between the peak transaortic pressure gradient and the frequency content of the murmur (r = 0.79) was demonstrated in a prospective "test" set of 50 patients with the clinical diagnosis of aortic stenosis. After heart sounds were recorded and digitized, three segments of the systolic murmur were isolated and analyzed by fast Fourier transform technique. An average frequency spectrum was quantitated by a previously described empiric spectral estimator. Clinical data and spectral ratio were correlated with the transaortic pressure gradient and aortic valve area was calculated from cardiac catheterization data. The best prediction of the transaortic pressure gradient was obtained when a 170 ms murmur segment was analyzed and when the predictive algorithm also included the aortic dimension (r = 0.87). The aortic valve area was poorly predicted (r = -0.48) unless estimates of blood flow and valvular calcification were included in the algorithm (r = 0.84). Further refinement of this technique may provide a non-invasive and clinically useful method for the estimation of aortic valve stenosis.