Mental workload is one of the most important factors in the performance of human and machine system, The reference for the cockpit design and the instrument verification can be provided by the research on the pilot's mental workload. The indexes of physiological measurement, performance measurement and subjective measurement were obtained by experiment. The mental workload sensitive index obtained by single factor analysis of variance, and the results showed that: the main effects of fixation frequency, total fixation time, blink rate, average pupil diameter, NASA_TLX and correct rate were significant (P<0.05).In this paper, combination research on GMDH and regression methods was used to establish the prediction model of pilots' mental workload, and the goodness of fit of the model was 85.47%. Therefore, this method can be used to predict the mental workload of pilots.