Abstract: Improved genetic programming is applied to predict rural labor migration. For improve speed and efficiency, simulated annealing is adopted to set up dynamically genetic operator probability. After the program trained by training samples, multi-dimensional prediction model of rural labor migration is established, and it is tested by test samples. The results shows that the function has good fitting and forecasting effect and can effectively avoid artificial error which is formed by various uncertain factors. Comparison with time series and general genetic programming prediction, the accuracy of improved genetic progamming prediction is 2.3 times for time series, and its run speed is 1/6 for general genetic programming. So rural labor migration prediction on improved genetic programming has favorable practical value.