Abstract: The paper aims to predict accurately the potato yield in order to draw on advantages and avoid disadvantages of climate. Based on the potato yield data in Datong of Shanxi in 1980-2015 and the climate data observed by national reference observatory in the same period, we established a prediction model of potato yield by using traditional statistical regression method and BP neural network method. The results showed that: firstly, using the method of the quadric curve and the least-squares, the climatic factors in potato sensitive stage were temperature, sunshine and precipitation, and precipitation had the greatest impact on potato yield; secondly, the improved climate yield algorithm could better reflect the function relation between climatic factors and crop yield; thirdly, the method of BP neural network, by the precision of training set to 0.005 and the learning rate set to 0.01 could approach the nonlinear function well on the Matlab platform; fourthly, more than 1/3 samples in the forecast verification showed that BP neural network model was better than traditional methods on prediction accuracy and fitting precision, and BP neural network had a very broad application prospect in potato yield forecast.