In this paper, different concentrations of trisodium phosphate solutions were used as research objects to explore the feasibility of UV/visible absorption spectra for detecting the concentration of trisodium phosphate solution in water. Three sets of absorbance measuring devices with different light sources and different optical path combinations were used to collect the spectral data of trisodium phosphate sample solution at different concentrations, and then the three groups of experimental data were pretreated by different pretreatment methods, the results of partial least squares (PLS) model showed that the best pretreatment method was Savitzky-Golay(S-G)convolution smoothing; then the method of Monte-Carlo cross validation (MCCV) was used to eliminate the singular samples in the calibration set, and the PLS model was established for the whole band data after the singular samples were removed, the results showed that MCCV method could improve the prediction performance of PLS models; finally, the main characteristic wavelengths of trisodium phosphate in three experiments were selected by correlation coefficient method, which were 214.099 8, 218.837 1 and 204.66 nm respectively, in order to establish a highly accurate and universal model for trisodium phosphate, the characteristic band of trisodium phosphate was selected from 200 to 222 nm based on the above three main characteristic wavelengths, and the corresponding characteristic band PLS models were established, the results showed that the model could be used quantitative analysis and prediction of trisodium phosphate concentration solution. Therefore, itis feasible to use hyperspectral technology to detect concentration of trisodium phosphate in water rapidly, and this study provides a theoretical basis for the development of instruments and equipment for the rapid detection of inorganic phosphorus concentration in water.