The spectroscopy sensing technology of water COD is an important development direction of modern environmental monitoring. Compared with traditional analytical methods, spectroscopy has more obvious advantages, such as continuous monitoring, online monitoring and fast testing, which is suitable for fixed-point and real-time monitoring of environmental water samples for COD. In this study, the ultraviolet absorption spectrum and the near infrared spectrum of real water samples were collected respectively by ultraviolet absorption spectrometry and near infrared transmission method. The COD prediction model was established by utilizing different spectral pretreatment methods combined with partial least squares regression(PLS) and multiple linear regression(MLR), and then the quantitative prediction and model parameters of ultraviolet and near infrared spectra measurement for COD were analyzed, finding that the Savitzky-Golay (SG) smoothing partial least-squares model had good prediction. Through comparison, the determination coefficients of prediction were 0.992 1 and 0.987 7, respectively, and RMSEP were 10.438 6 and 5.972 0, respectively. Ultraviolet and Near-infrared spectroscopy combined with MLR analysis model had poor prediction, with the determination coefficients of prediction 0.928 0 and 0.957 3, respectively. Through a comprehensive analysis of the experimental results, ultraviolet absorption spectrum prediction model in 280～310 nm spectral region had a good performance. Near infrared spectral spectrum model had the best performance in 7 250～6 870 cm-1 spectral region. Ultraviolet spectrum corresponding to the decision of prediction model was higher, but the near spectrum model had better stability and repeatability. Studies show that the spectrum sensing technology can be used in the quantitative predicted analysis of COD in actual water. The conclusion from the paper laid a theoretical basis for the development of portable water testing equipments.