Abstract: The average-velocity method is the most commonly-used time-to-depth conversion technology for the seismic survey of coalbed. But for the steeply-inclined coalbed, the average velocity changes rapidly laterally, and the interpretation accuracy varies greatly, so that the intelligent coal mining could not be assured effectively. Herein, the accuracy of various average-velocity-interpolation methods was discussed by analyzing a steeply-inclined coalbed model and a practical 3D seismic mining area and appropriate improvement methods were proposed accordingly. The calculation results of a forward model show that the bottom elevation accuracy of coalbed is greatly affected by the locations of interpolation points in the case of direct interpolation using Kriging and Polynomial methods, which cannot meet the high-accuracy requirements of time-to-depth conversion for steeply-inclined coalbeds. Therefore, referring to the stack-velocity derived average velocity, an improved interpolation method of time-to-depth conversion applicable to the steeply-inclined coalbed was proposed by integrating the polynomial, Kriging and Support Vector Machine (SVM) methods, with reference to the average velocity calculated by stacking velocity. By applying the relevant methods to the 2D model data and the 3D example of mining area, it is found that the accuracy of average velocity generated by the improved interpolation method is significantly improved, with the interpolation error much smaller than the requirement of the industrial standard. Hence, the proposed method is suitable for large-scale popularization and application.