The purpose of this study is to analyze human pilot maneuvers during the landing phase of visual approach. The pilot model is constructed by using Neural Network, and is trained with the time history of flight data recorded in flight simulation or real flight. The inputs to this model are visual cues and the memory of column operations. The outputs are the column deflection and the engine throttle position. A Genetic Algorithm approach is proposed to improve the generalization capability of the network by determining the network structure and the initial values of weights and biases before training. Finally, the characteristics of each pilot's skill is revealed by Contribution Ratio Analysis and Sensitivity Analysis for the obtained neural network model. The method was applied to the analysis of real flight.