AbstractThe discrimination between four types of upper limb movements, namely: flexion, extension, pronation, and supination, from the EMG patterns of surface electrodes located at the biceps and triceps of the upper arm is treated. To discriminate between these limb movements, four different models are utilized to assess their potential for achieving high discrimination rate. The four models are: autoregressive (AR), integrated autoregressive (IAR), integrated autoregressive derived from an overdetermined set of equations, and finally an integrated autoregres-% sive that emphasizes the lower frequency part of the EMG, spectrum. The AR approach does not produce stable model in several cases which indicates the inadequency of this approach for EMG signal due to its nonstationary characteristic. The other models yield stable systems and the parameters of their models are utilized to discriminate between the aforementioned movanents using linear discriminant function approach. Experimental results are presented to demonstrate the success rates of each of those models in recognizing the movement function performed.