Three experiments were carried out to evaluate the relationship between two techniques for quantifying electroencephalographic (EEG) data during sleep: period amplitude analysis (PAA) and power spectral analysis (PSA). In Experiment 1, canonical correlations and regression analyses were computed on PSA and PAA data from 40 undergraduate volunteers. The results yielded an average canonical correlation of 0.98. Further, multiple regression analyses demonstrated that the PSA variables accounted for approximately 66% of the variance in the PAA data, whereas PAA variables captured 88% of the variance in the PSA data. Epoch-to-epoch correlations were higher for PAA measures than for PSA data, perhaps indicating greater stability of PSA measures across epochs of sleep. In Experiment 2, PSA and PAA data were compared in 17 unmedicated outpatients with unipolar depression. Canonical correlations and regression analyses indicated that the overlap in variance between PSA and PAA did not exceed 50%, regardless of whether PSA or PAA variables were used as predictors. Epoch-to-epoch correlations between PAA measures were significantly higher than correlations among PSA variables, again suggesting greater stability of PAA data across epochs of sleep. The range of correlations for either data set was, however, substantially lower in the depressed than in the normal group. Experiment 3 evaluated the possibility that filter settings and artifact-rejection procedures had contributed to reduced overlap in PSA-PAA variance and reduced stability in depressed patients. An additional group of eight healthy volunteers served as subjects. Findings in Experiment 3 indicated that methodological differences between Experiments 1 and 2 did not account for the reduced correlations in the depressed group. It was concluded that PSA and PAA data should be comparable in normal subjects but are relatively independent in depressed patients. Epoch-to-epoch correlations were higher for PAA data than those found between PSA measures in both normal subjects and depressed patients, suggesting that PAA may be more stable across sleep epochs. Reduced stability may be a reflection of nonstationarity in the EEG of depressed patients.