Animal studies have shown that the sleep-related oscillations in the frequency range of spindles and slow-waves, and in the gamma band occur synchronously over large parts of the cerebral cortex. Coherence analysis was used to investigate these oscillations in the human sleep electroencephalogram. In all-night electroencephalogram recordings from eight young subjects power and coherence spectra within and between cerebral hemispheres were computed from bipolar derivations placed bilaterally along the antero-posterior axis. The 0.75-50 Hz range was examined with a resolution of 0.25 Hz. Distinct peaks in coherence were present in non-rapid eye movement sleep but not in rapid eye movement sleep. The most prominent and consistent peak was seen in the range of sleep spindles (13-14 Hz), and additional peaks were present in the alpha band (9-10 Hz) and low delta band (1-2 Hz). Whereas coherence in the spindle range was highest in stage 2, the alpha peak was most prominent in slow-wave sleep (stages 3 and 4). Interhemispheric coherence at 30 Hz was higher in rapid eye movement sleep than in non-rapid eye movement sleep. There were also marked sleep state-independent regional differences. Coherence between homologous interhemispheric derivations was high in the low frequency range and declined with increasing frequencies, whereas coherence of intrahemispheric and non-homologous interhemispheric derivations was at a low level throughout the spectra. It is concluded that coherence analysis may provide insights into large-scale functional connectivities of brain regions during sleep. The high coherence of sleep spindles is an indication for their widespread and quasi-synchronous occurrence throughout the cortex and may point to their specific role in the sleep process.