The full spectral analysis (FSA) technique is investigated using both experimental and model-generated data. The relationship between the formulae presented by different authors is discussed, and the similarities and differences inherent in using these formulae are evaluated. The FSA is applied both to sample spectra calculated using individual time series records and to spectra produced by the incoherent average of the sample spectra. The results reveal that the velocities obtained using sample spectra can be significantly biased in the situation where large turbulent motions exist for two particular implementations of the technique. These biases are shown to decrease as the number of sample spectra used in the incoherent average are increased, thereby producing FSA velocity estimates in excellent agreement with those produced using the full correlation analysis (FCA) technique.¿ 1997 American Geophysical Union