Abstract: In order to improve efficiency and accuracy of gas emission prediction, we proposed gas emission prediction new method combining extreme learning machine (ELM) and genetic algorithm (GA). In order to avoid ELM predicting effect affected by the random of the input layer weight matrix and the hidden layer bias, GA was used to optimize the input layer weight matrix and the hidden layer bias, and GA-ELM gas emission prediction model was built. Case analysis was made using statistical data of a coal mine, and the prediction result was compared with ELM, SVM and BP. The results showed that the GA-ELM model can relatively accurately and efficiently predict the gas emission.