An intuitive control system based on gesture recognition for humanoid robot has been developed. The gesture of a human operator can be recognized successfully with 99.87% accuracy using Fuzzy Neural Generalized Learning Vector Quantization (FNGLVQ) algorithm. A total of 13 distinct gestures were trained onto the system to represent 13 different actions. The system maps the recognized gestures to 13 different sequences of movements which will be executed by the humanoid robot. This system can be used in settings where tele-operation of humanoid robot is needed. The use of gestures to give commands to humanoid robot will decrease the learning curve of the operator by providing a more intuitive control mechanism. This system eliminate the need to memorize a set of complex procedures. Another possible application is for teaching children with autism basic social interaction as a complement for human tutor.