Abstract: A method based on the neural network for the forward kinematics solution of Stewart platform is studied. The training examples are obtained through the inverse kinematics, and then preprocessed. The final neural network includes six sub-networks, corresponding to the six degrees-of-freedom. Unfavorable error distributions in the network are improved through re-construction of the training examples. The simulation results show that the neural network operates well and provides the forward kinematics solution an error of 0.5° and 3 mm when the platform runs in its full range.