Abstract: Five seismic parameters such as amplitude of wave crest and hollow(A¹),average frequency(Fa),energy in dominant frequency domain(Qf¹),energy in low frequency domain(Qf),peak frequency(Fmain) are derived according to the seismic kinematics and dynamic characteristics of coal layer thickness.Eight groups of studying samples,made use of BP(Back Propagation)neural network of four layers improved by adopting momentum algorithm and self-adaptive-adjusting learning rate algorithm to train the BP neural network,and used the trained BP network to forecast coal layer thickness.It was proved that forecasting coal layer thickness by BP neural network from multiple seismic parameters had high accuracy by the practical data.and is an effective approach for forecasting coal layer thickness.