Abstract: After raw coal gas was extracted from coal seam,the danger of gas outburst cannot be eliminated completely. Therefore,the occurrence law of residual gas after extraction and its prediction must be studied. The geological con-ditions of mining area No. 3233 in Xutuan coal mine in the Huaibei coal field,China,were analyzed. The structural in-dex and its distribution in the study area were evaluated according to the calculation and statistics of fractal dimension of faults,the structural curvature of coal seam floor and dip angle of coal seam,and their giving weights of 0. 35,0. 35, and 0. 30. The influence of structural index,coal seam depth,coal thickness and raw coal gas content on residual gas after extraction was analyzed by using 42 sets of selected data. The correlation between the loss of gas content (LGC) and influencing factors,e. g. ,structural index,coal seam depth,and coal thickness,was fitted through multiple linear regressions. The feasibility of predicting LGC after extraction was studied by using the back propagation (BP) artificial neural network model. Results show that the structural index can quantitatively and accurately characterize the struc-tural complexity of the mine. The LGC is mainly influenced by structural index,the buried depth of coal seam,coal thickness,and raw coal gas content. In general,LGC is negatively correlated with structural index and positively corre-lated with other influencing factors. According to the F-test of mathematical statistics,i. e. ,F = 20. 82>F0. 01(3,38)= 4. 35,the results of multivariate linear regression is significant,indicating a close internal relationship between the LGC and the indexes of the influencing factors. Among these aspects,the structural index has the greatest influence on LGC,whereas the burial depth of coal seam has the least influence. The influence of coal thickness is between that of structural index and the burial depth of coal seam. A BP artificial neural network model with 4×10×1 structure was es-tablished with LGC as the output index and the structural index,the buried depth of coal seam,coal thickness and raw gas content as the input indicators. The accuracy of model prediction results is high after learning and training,and the relative error is 1. 19% -1. 34% ,indicating that the artificial neural network model can be used to predict LGC after extraction in the unmined area. Residual gas content is the original coal gas content minus the LGC. Hence,this value can indirectly predict the residual gas content after extraction.