Abstract: For the purpose of predicting the loss of surface elevation caused by coal mining, the traditional poor-information-based GM(1,1) model, GM(1,1) residual model and timing residual GM(1,1) model were set up according to grey system modeling, correlation analysis and residual recognition. Meanwhile, the original-data-based grey CompertzI model with absolute error, the grey LogisticI model, and the grey CompertzII model and grey LogisticII model that both have relative error were also set up. Then, these models were applied to the actual prediction analysis of subsidence value in Tuzhu Coal Mine of Jinzhushan Mining Company. After a grey generation was made on the loss of surface elevation from January to October in 2007, seven grey prediction models were set up. Comparative analysis on the accuracy test of these prediction values gained by these grey models shows that the seven models are all firs-level (good) models, and that the grey CompertzII model and grey LogisticII model, due to having a high prediction accuracy and reliability, are far superior to the traditional GM(1,1) model. They play an important role in the restoration planning of coal mining.