Abstract: This paper described a fuzzy neural network(FNN) used for identification of gangues. The niche genetic algorithm(NGA)based on a proposed practical niche operator was responsible for the off line training of FNN. Furthermore, the test results showed that not only the off line trained FNN employing NGA had a satisfying global convergence, but also the fuzzy rules from the resultant quantitative FNN provided the separator with a higher recognition accuracy.