Organic pollutants are the main source of water pollution. The degree of organic pollution in water bodies can be comprehensively reflected by chemical oxygen demand (COD) indicator. Compared with the traditional methods, which are highly time-consuming and cause secondary pollution, remote sensing techniques can detect water COD quickly and effectively, particularly over large areas. However, due to complex composition of COD, detecting water CODusing remote sensing technology is still insufficient, and the spectral response mechanism of COD in the visible-short-wave infrared range is not yet clear. To clarify the spectral response mechanism of water COD, the reflectance spectra of 45 different concentrations of COD standard solution (potassium hydrogen phthalate) were measured by using PSR hyperspectral instrument in the laboratory simulation environment in this study. The continuum removal and reflectance normalization methods were used to analyze the spectral characteristics of COD standard solution with different concentrations, and results showed that with the increasing of COD concentration, the water reflectance spectra increased gradually in the visible-short-wave infrared range, as well as the spectral response increased rapidly in the range of 540～580 and 1 000～1 060 nm, showing the three-stage frequency doubling of —OH stretching vibration and the combined frequency absorption characteristics of —CH stretching vibration and deformation vibration. In order to further verify the effectiveness of sensitive bands, the partial least squares (PLS) regression models were developed by using sensitive bands and full bands, respectively. The correlation coefficient and RMSE of COD standard solution based on the sensitive bands were 0.972 and 39.629 mg·L-1, respectively, while the correlation coefficient and RMSE based on full bands were 0.961 and 46.639 mg·L-1, respectively. It showed that for COD standard solution that is not interfered byexternal factors, high accuracy can be achieved with only a small number of sensitive bands, even higher than the full bands model. The model retrieval accuracy based on the sensitive bands was also significantly better than the accuracy based on full bands when the model are applied to the field-measured water spectra. The results suggested that using the 540～580 and 1 000～1 060 nm sensitive bands can effectively improve the accuracy of detecting water COD, and significantly advance our ability for large-area rapid monitoring of COD using remote sensing in practical cases.