An error analysis of the one-bit autocorrelation method of spectral estimation is presented, and the variance on the autocorrelation estimate and spectral estimate is given. In the nonclipped (many-bit) case, the normal Nyquist sampling rate of 2B is sufficient. Higher sampling rates do not further improve the spectral estimate. In the clipped (one-bit) case, increasing the sampling rate does improve the estimate. The present work was initiated to examine this effect. The results verify the decrease in spectral variance with increased sampling rate. Itis shown that a part of the signal-to-noise deterioration caused by the clipping may be removed by the higher sampling. An increased sampling rate from 2B to 4B, however, achieves most of the available improvement.