Abstract: Identification of mine water inrush source is of great significance in mine water hazard control. The six index of Ca2+、Na++K+、Mg2+、HCO3－、Cl－、SO42- were selected as variables water samples. Combining principal component analysis and Bayesian discriminant simplification to build water inrush identification model. The water quality analysis data of the Cenozoic loose beds, coal-serial sandstone and Taiyuan Formation limestones in Panji No.2 Mine were used as training samples and prediction samples, including 24 training samples and 11 prediction samples. The results show that the correct rate of water in the loose layer is 81.8%, the correct rate of sandstone water is 83.3%, the correct rate of limestone water is 85.7%, the overall correct rate is 83.3%, and the reliability of the discriminant results is high. At the same time, the principal component analysis and Bayesian combination water inrush recognition model compared with Bayesian model shows that the combination of principal component analysis and Bayesian model can effectively eliminate redundant information and make the discriminant results more rapid and accurate.