Backscattered ultrasonic waveforms from selected regions of the liver were collected from B-scans of 11 male patients ranging from 13 to 41 years of age and analyzed to determine local values of both the magnitude and frequency dependence of attenuation. Processing was based on frequency domain analysis and also incorporated precise corrections for time-varying gain, nonlinear amplifier compression, and beam diffraction, which would otherwise degrade accuracy. The results demonstrate that (a) consistent and reproducible measurements of attenuation from one scan to the next are possible within a given patient, and (b) frequency dependence can deviate significantly from the linear relationship between frequency and attenuation commonly assumed for soft tissue. Based on the accuracy of the overall analysis and the reproducibility of the results, the authors suggest that a multivariate approach to clinical tissue characterization using both the magnitude and frequency dependence of ultrasonic attenuation may be possible.