This article describes the design of a microcontrolled bioinstrumentation system for active control of leg prostheses, using 4-channel electromyographic signal (EMG) detection and a single-channel electrogoniometer. The system is part of a control and instrumentation architecture in which a master processor controls the tasks of slave microcontrollers, through a RS-485 interface. Several signal processing methods are integrated in the system, for feature extraction (Recursive Least Squares), feature projection (Self Organizing Maps), and pattern classification (Levenberg-Marquardt Neural Network). The acquisition of EMG signals and additional mechanical information could help improving the precision in the control of leg prostheses.