Fast Fourier transform (FFT)-based coherence estimation is widely used for investigating neuromuscular coupling in tremor. The significant coherence was often found not only at the tremor frequency but the harmonic frequencies as well. It is necessary to differentiate the genuine physiological component from the harmonics. We investigated the factors contributing to generation of the significant coherence estimates at the harmonic frequencies using simulated signals and physiological signals of electromyographys (EMGs) of the forearm extensor/flexor and oscillatory local field potentials (LFPs) of the subthalamic nucleus (STN) in Parkinsonian tremor. Rhythmic pulses varied in duration simulating the envelope signal in tremor EMGs and sine waves varied in waveform and level/frequency distribution of added noise simulating the tremor signals in the STN LFPs were formed. With ethics committee approval, tremor-related EMGs and STN LFPs were recorded from the Parkinsonian patients following deep brain stimulation surgery. Coherence estimation was computed between the simulated and physiological signals. The effects of pulse duration, waveform, noise level and frequency distribution were compared on the coherence results of a range of paired signals. Our results of the simulated signals showed that (1) although the signal of rhythmic pulses was significantly distorted from the perfect sine wave, which generates harmonics on the power spectrum (the shorter the duration, the more harmonic peaks appeared with lower power). This did not significantly influence the amplitude of harmonic component in the coherence estimation; (2) in comparison with near perfect sine wave, the half-wave simulated signal significantly increased the harmonic component in both number and coherence value; and (3) added white-noise significantly decreased the coherence value of harmonics, whilst the noise distributed in a particular frequency range selectively decreased the coherence value in the corresponding frequency range. These results suggest that the complexity of the simulated signal was the most significant factor for generating the harmonic peaks in the coherence estimation. Our results of the coherence estimation of tremor-related EMGs and STN LFPs obtained from different patients confirmed that (1) harmonic peaks are likely to appear when the tremor-related component is the most dominant one in the STN LFPs; and (2) the non-tremor related component in the STN LFPs may suppress the coherence value selectively depending on its frequency distribution. We conclude that the harmonic peaks in the coherence estimation of tremor-related neuromuscular signals are related to the degree of distortion in waveform of the tremor-related component and significantly modulated by the non tremor-related component.