Paired-pulse transcranial magnetic stimulation (TMS) is used to measure the excitability of interhemispheric inhibition (IHI) between the hand areas of the two motor cortices. It varies from person to person, and is highly predictive of individual differences in callosal anatomy (fractional anisotropy) and even motor behaviour, e.g. the amount of involuntary electromyographic (EMG) ‘mirroring’ in one hand during rapid contraction of the other. The present experiments tested whether it also predicts how well individuals can improve motor performance in a task involving the two hands. Healthy participants were given 100 trials to maximize the initial acceleration of a ballistic finger movement made with one hand while trying to maintain a tonic low level of EMG activity in the other hand. Initially, each movement was accompanied by additional unwanted EMG mirroring in the other hand. However, after practice, participants had on average increased acceleration by approximately one-third without changing the amount of EMG mirroring in the contralateral hand; indeed, in some individuals EMG mirroring activity declined. TMS measures showed that there was an increase in corticospinal excitability in the trained hemisphere, but there was no change in the excitability of short- or long-latency IHI from the trained to non-trained hemisphere. Nevertheless, in each individual, the baseline (pre-practice) excitability of short-latency IHI was highly predictive (r = 0.65; P = 0.0019) of the change in EMG mirroring. The implication is that a physiological measure of brain excitability at rest can predict behaviour in response to training.