Abstract: For problems of existing discrimination methods of water inrush source based on hydrochemical characteristics such as complex process, difficulty to distinguish single water inrush point, neglecting the correlation among hydrochemical characteristics, large calculation amount and so on, a discrimination method of mine water inrush source based on Bayes-extension discrimination method was proposed combining with Bayes discrimination method and extension discrimination method. Twenty-six groups of water samples from different aquifers and four groups of water samples to be judged in Pa'er Coal Mine are obtained, the content of SO2-4, Cl-, HCO-3, K++Na+, Mg2+, Ca2+ in the water samples is taken as hydrochemical characteristic indexes, and thus matter-element models of the water samples are established. The correlation degrees between the water samples to be judged and the known water samples are obtained by use of extension discriminant method. The miscalculation loss of Bayes discriminant method and density function of the water samples to be judged with the known water samples are combined to get Bayes-extension solutions. The type of water inrush sample is discriminated according to the minimum value of Bayes-extension solutions. Piper trilinear diagram, extension discriminant method and Bayes-extension discriminant method are adopted separately to discriminate the type of mine water inrush samples. The results show that Piper trilinear diagram is difficult to accurately discriminate a certain type of water samples, extension discriminant method has misjudgment, while Bayes-extension discriminant method can accurately discriminate the type of water inrush sample.