Based on the principle of molecule fluorescence of rhodamine 6G, the fluorescence spectra under different experimental conditions are compared and the maximum fluorescence intensity is obtained when pH is 1. When molybdate, potassium dihydrogen phosphate and sulfuric acid are added into the rhodamine 6G reagent the complex is generated and the fluorescence intensity of rhodamine 6G declines. Within a certain range, it exhibits linear relationship. The position of fluorescence peak does not change. A nonlinear model is constructed based on genetic algorithm-back propagation (GA-BP) neural network which consists of a 36×18 matrix as inputs and a 1 × 18 matrix as outputs, and its purpose is to detect the phosphate concentration. In network training, the error accuracy is 10-3 and the correlation coefficient between the outputs and the expectations is 0.998. In network prediction, the average recovery is 99%, while the average standard deviation is 1.79% , reaching the ideal results. Therefore, this network can better detect phosphate concentration of 0~2.00 mg/L. In summary, a quick and effective way to detect phosphate concentration is provided, which helps promote the development and application of environmental monitoring technique.