Abstract: It is valuable to exploit the coalbed-methane which is rich in our country. The prediction of gas content in coalbed methane reservoir is a key step in the early stage of development and utilization. In recent years, BP neural network algorithm has been often used in coalbed methane prediction, but the model has some shortcomings in the training process, such as slow convergence speed, sensitive to initial value and easy to fall into local minimum value. Therefore, this paper proposed an improved BP neural network prediction model characterized by artificial bee colony algorithm. Firstly, R-type cluster analysis was used to classify the seismic attributes extracted from the 3D seismic data, four seismic attributes which are most sensitive to the change of coalbed-methane and independent of each other were selected. Secondly, the artificial bee colony algorithm(ABC) was used to find the optimal connection weight of the input layer and the hidden layer and the optimal threshold of the hidden layer of BP neural network, to build a robust ABC-BP neural network prediction model, and the seismic attributes of well location and gas content data was used as samples to train the model. Finally, the coal seam gas content in the work area was predicted by taking the optimal seismic attributes of the target reservoir in the whole work area as input. The prediction results are basically consistent with the change trend of gas content in each well. Among them, the average error rate at the training well is 0.23%, and the error rate at the verification well is less than 15%. Therefore, the prediction method has high reliability and strong applicability, and can be effectively used for coal seam gas content prediction.