Abstract: Water inrush is one of the major disasters in coal mining, water inrush occurs, rapidly and accurately judging water inrush water is vital for flood control. When identifying the mixed water, Fisher discriminant analysis method can only confirm the most similar group, resulting in miscalculation, so on the basis of Fisher's discriminant analysis theory the centroid distance evaluation method was introduced to eliminate possible mixed water sample data, it can improve the accuracy of determination results. For this purpose, 13 kinds of hydrochemistry ions were selected as a discriminant factor, 30 representative samples as training sample to establish discriminant function model, using the discriminant function model and evaluation index of center of mass, six single water samples taken from specific aquifers and 4 mixed water samples from sump were verified and analized, after introduction of centroid distance evaluation and analysis, the discriminant accuracy increased from 60% to 83.3%. The research results show that using the centroid distance evaluation analysis after Fisher discriminant analysis is very useful to improve the accuracy of determination results.