Abstract: The architectures and training algorithms are complicated when BP and RBF neural networks are used in the sources recognition of coal mines water inrush.To overcome this problem, perceptron is used to discriminate the type of water inrush.The ion concentrations of Na⁺ and K⁺ are omitted on purpose and those of Ca²⁺、Mg²⁺、Cl^-、SO^2-₄ and HCO^-₃ are selected as the basis of water sources recognition.A perceptron model with six inputs and four outputs is established by using 35 water samples from the Jiaozuo mine area.The computation results show that the perceptron is an effective recognition tool which can identify the sources of water inrush correctly with different learning rates and initial weighting matrices.