The mine water source discrimination plays an important role in guiding mine water prevention in the water prevention work. Improving especially the discrimination accuracy of mine water sources which will cause water inrush is the important foundation of avoiding such accident and thus ensuring personnel and property being in safety. Based on Fisher discriminant analysis theory (FDA) and gray correlation analysis theory (GCA), groundwater chemical components (Ca2+, Mg2+, K++Na+, Cl−, SO4 2−, and HCO3 −) data at main water inrush aquifers in a typical coal mine through experiments, FDA-GCA recognition model of water inrush sources was established and then verified. Results indicated that the FDA-GCA recognition model of water inrush sources was characterized by high discrimination precision, and false determination rate obtained by back-substitution estimation method was zero. Thus, it had strong discrimination ability of water inrush sources. The significance ranking of input variables in the water source FDA-GCA recognition model was Ca2+ > Mg2+ > HCO3 − > SO4 2− > K+ + Na+ > Cl−. The data in this paper were discriminated by the distance discriminant method and BP neural network discriminant analysis method, the correct rate of which is 90%, 80%, slightly lower than or equal to the accuracy of 90% by the FDA-GCA recognition model.