This investigation explored pattern recognition of EMG signals produced by shoulder area muscles to identify the performance of a select number of lower arm motions. The signals were modelled as fourth-order autoregressive processes with the parameters of the models used to classify the different motions. The recorded EMG signal was bandpass filtered for each individual to improve discrimination between signals. The method was shown to detect and identify EMG signatures produced by at least two, and sometimes three, different arm motions. Discrimination between the signal models for the motions was not affected by load variation or muscle fatigue.