Variations in the amplitude and period of essential and Parkinsonian tremors were studied. It was found that the variations in frequency (or period) were mostly similar to the white noise, with the standard deviation typically less than 10% of the mean, whereas the variations in amplitude were much larger, with standard deviations greater than 30% of the mean, and so could not be effectively smoothed by running means. It is conjectured that variations in frequency reflect the stable nature of the neural network that generates the rhythmicity responsible for the tremor. The variations in amplitude, however, reflect more the fluctuations in the firing of individual neurons in the network. It is further discussed that the oscillator behind pathological tremor has a stochastic nature and can be characterised as a diffusional process. The latter suggests that it is sometimes possible for tremors to be described as chaotic processes on certain scales in phase space. It is further discussed how the stochastic nature of tremors determines the lack of correlation between different tremulous parts of the body.