Data of the UV absorption spectra from the water samples with a certain concentration of the nitrogen and phosphorus is very large. Wavelet transformation (WT), based on the soft-threshold, can be effectively used to compress the spectrum signal of UV absorption spectra. High correlation often exists among the spectral intensity for the water samples with different concentration of the nitrogen and phosphorus. Partial least square regression (PLSR) can be used to decrease such correlation and to build an effective regression model for the sampled water. The detection precision is also increased in this way. Simulation of the tested water samples showed that the maximum relative error in the concentrations of nitrogen and phosphorus was 8.9%. The model can fully meet the requirements of detection precision.