Before landslides slide partly, the displacement of part monitoring points in the landslide will change according to a discipline, when landslides slide partly, these monitoring points will cause a great displacement in a very short time, it is called as a sudden change displacement, after landslides slide partly, the displacement of these monitoring points will change according to the other discipline. So we can use the definition of “+” function to erect the uniform landslide deformation forecast model.To different landslides and monitoring points at different locations of landslides, because their geographical environment and geological environment are different, their deformation discipline is different, we can preplace some deformation models and let the computer look for the deformation model whose forecast error is least. In order to improve the fitted accuracy and forecast accuracy of the forecast model ulteriorly, parameters of the deformation model whose forecast error is least are looked as status vectors with dynamic noises, Kalman filter model is erected to forecast the deformation of the landslide. Examples of monitoring point B6 and B7 in a landslide verify that the fitted accuracy and forecast accuracy of Kalman filter model is good, the method can be used in the forecast of landslides.