Function discrimination via EMG parameter identification has been employed for controlling electrical stimulation of paralyzed nerves to provide paraplegics with certain walking functions. In this case, large amounts of voluntary EMG data from muscles above the level of paralysis have to be processed in order to discriminate four basic functions needed by the patient to perform the desired walking functions of sitting down, standing up, lifting the right leg and lifting the left leg. The purpose of this study is to extract from the observed data more significant information that may represent the basic functions and to discard less significant data. This will make the classification of functions more efficient. The authors discuss the Karhunen-Loéve expansion and its application to this feature, which motivates an algorithm.