A design method for neural network (NN) based feedforward controller in the framework of neurointerface is proposed for nonholonomic robots by applying a concept of virtual master-slave system, in which a master robot is assumed to have a stable inverse dynamical model that includes unknown physical parameters. A PD-based feedback compensator is simply added to the neurointerface to suppress an output deviation caused by the mapping error of NN. The effectiveness of the present approach is shown by a simulation for a tracking control problem of a nonholonomic mobile robot with two-independent driving wheels.