We present the improved control for a neural prosthesis (NP) for functional electrical therapy of standing and walking of hemiplegic individuals. The rule-based control is based on a finite state model that uses kinematics of the legs as input and stimulation of muscles as the output. The rules were determined by an artificial neural network (ANN) through the off-line training that used kinematic data and EMG recordings from eight healthy individuals while standing and walking. The validity of the mappings was evaluated by a questionnaire. The same healthy individuals exposed to stimulation graded their appreciation of the timing of the stimulation through four answers. Results suggest that the timing is applicable for activation of muscles; yet, that the stimulation sequences adjusted for up to 5% meet individual needs better. This finding suggested that the If-Then rules determined by ANN could be used as the initial stimulation sequence in control of a NP for standing and walking. The implementation of this control is being tested in a clinical study that includes 16 hemiplegic individuals. Here we present only the results that relate to the recovery of balance, since the follow up of other outcome measures is still in progress. The findings are that adding the NP augmented treatment of standing leads to significantly better balance recovery compared with the conventional therapy only