The variability of the amplitude of the motor evoked potential following transcranial magnetic stimulation is well known, however the mechanisms underlying this variability have yet to be characterised. As a first step in developing models to explain this variability the statistical properties of have to be precisely characterised. The aims of this study were to characterise the distribution of the motor evoked potential following magnetic stimulation. Healthy volunteers were used (2 female, 3 male, aged between 22 and 37). Magnetic stimulation was produced using a Magstim Rapid (The Magstim Co. Ltd, UK), motor evoked potentials were recorded from abductor pollicis brevis muscle of the right hand, using a belly/tendon method, with a Dantec Keypoint Electromyography Machine (Medtronic functional Diagnostics, Denmark). Stimulation was performed at 50%, 70% and 90% of the maximum stimulus intensity. For each subject, 20 trials were performed at 50% and 90% and 50 trials at 70%, as well as 20 trials at 70% with the target muscle activated. To examine the variation of responses across stimulation levels and subjects, a two way analysis of variance was performed. The residual variability of responses after subtracting effects of stimulation levels and subjects was then examined. 90% of the variability in responses was due to subject or stimulation level effects or their interactions. The remaining 10% is residual variability of the MEP within individuals at a given stimulation level. The distribution of this residual variation (X) appears non-normal, with a kurtosis value of 7.91 (for the normal distribution, kurtosis = 0). A power law transformation (Y = X0.3), calculated via the Box-Cox method, appears to give best fit to the normal distribution. There is substantial variability in MEP levels across subjects and stimulation levels. In addition, the residual MEP response is not normally distributed. The non-normality has implications for the summary statistics used to describe the variability of the MEP response. Any models developed to account for the variability of these responses should explain this non-normal distribution.