It is well known that computed torque robot control is subjected toperformance degradation due to uncertainties in robot model, and application ofneural network (NN) compensation techniques are promising. In this paper weexamine the effectiveness of neural network (NN) as a compensator for thecomplex problem of Cartesian space control. In particular we examine thedifferences in system performance of accurate position control when the same NNcompensator is applied at different locations in the controller structure. It isfound that using NN to modify the reference trajectory to compensate for modeluncertainties is much more effective than the traditional approach of modifyingcontrol input or joint torque/force. To facilitate the analysis, a new NNtraining signal is introduced and used for all cases. The study is also extendedto non-model based Cartesian control problems. Simulation results withthree-link rotary robot are presented and performances of different compensatinglocations are compared.