Abstract: The known coal thickness points are mainly distributed in “U” type roadway for coal mine working face, such as tail gate, head gate and starting cut. Because of the coal thickness points’ uneven density, the commonly used interpolation method of seam thickness is difficult to be used except the Kriging space valuation techniques. Due to accurate prediction of coal thickness variation is very important for efficient mining of coal mining face,based on the “U” type’s coal thickness “static data” to establish the initial model, it can be done for coal thickness prediction in first stage; with the development of working face, the initial model can be updated and optimized because more and more new coal thickness points is added during the mining stage, so the progressive prediction of coal thickness can be done for the unmined area.The No.S1 working face is selected as an example which has 306 known coal thickness points in “U” type roadway and 204 coal thickness points exposed by mining as the verification points. Taking 50 m as a mining stage, S1 working face can be divided into 29 mining stage. The result of precision test and error analysis for coal thickness progressive prediction include:①On the basis of the known points in “U” type roadway to establish a static prediction model, when the coal thickness absolute prediction error is less than 0.10、0.50、1.00 m, the rates of measurement points in working face are 25.98%, 74.50%, 90.73% in S1 working face.②For the progressive prediction model by collaborative coal thickness data from “U” type roadway and new data by mining, when the coal thickness absolute prediction error is less than 0.10、0.50、1.00 m, the percent of measurement points in working face are 48.01%, 85.51%, 94.86% respectively. It had shown that the coal thickness prediction accuracy was greatly improved by progressive prediction.