This study reports the findings of an analysis of temporal correlation between tremor of different muscles of the same and different limbs in four patients with Parkinson's disease. Spectral coherence methods were used for determining whether simultaneously occurring oscillations in the electromyograms of different muscles are statistically coupled. The incidence of significant coherence was considerably higher for muscle pairs in the same limb than for pairs in different limbs; Parkinson's disease tremor is coupled within but not between limbs. Because the characteristics of tremor are known to vary under different behavioral situations, the intralimb coupling was examined for different tasks. A mental arithmetic task resulted in an increase in the coherence between muscles of the same limb, whereas the finger-to-nose task decreased the coherence. No significant change in coherence was found for a postural task. The amplitude and regularity of tremor electromyography showed changes analogous to those in coherence. These results support the hypothesis that tremor in different limbs results from the activity of several neural circuits oscillating independently. The results also emphasize the value of these methods for rigorously characterizing tremor, in relation to disease state, behavioral conditions, and the selection of treatment strategies.