Spectral analysis of heart rate variability (HRV) is useful as a noninvasive means of assessing autonomic function in patients with obstructive sleep apnea (OSA). However, standard spectral measures, such as the ratio of low-frequency to high-frequency power (LHR) and normalized high-frequency power (NHFP), can be confounded by the abnormal breathing patterns that occur during sleep. To circumvent this limitation, we employed an autoregressive modeling approach to partition the RR time-series into a component that is correlated with respiration and a respiration-independent component. From these components, we derived two new spectral indices: the modified LHR (MLHR) and the average gain relating respiration to RR changes (GRSA). Six normals and seven OSA patients were studied in relaxed wakefulness and stage 2 sleep; during sleep, the OSA patients were studied without and with continuous positive airway pressure (CPAP) therapy. All four spectral indices showed significant differences between OSA patients and normals in both wakefulness and sleep, although the changes in MLHR and GRSA were substantially larger and less variable: MLHR (p < 0.0003) and GRSA (p < 0.0001) vs. LHR (p < 0.005) and NHFP (p < 0.004). However, in the OSA subjects, LHR and NHFP were unchanged by CPAP. By contrast, CPAP produced a highly significant increase in GRSA (p < 0.0004), as well as a decrease in MLHR (p < 0.03). Thus, by compensating for the effects of breathing pattern differences, MLHR and GRSA unmasked the effects of CPAP therapy, which has been shown in previous studies to reduce sympathetic activity and increase vagal cardiac modulation.