Chemical oxygen demand (COD) is one of the important indicators of water quality. The COD measurement method based on UV/Vis absorption spectra has been widely used because of its advantages of high speed, real time results, void of reagents, and pollution-free. Different functional groups have different characteristic absorption spectra. In the application of the method to water samples with a stable solution, high COD measurement accuracy can be achieved by building a precise “UV/Vis Absorbance(Auv)-COD” computational model. However, when the method is used for water samples with variable components, the measurement precision is low, limiting its applicability. This study proposes a new method based on dynamic identification of water sample type. The LM-BP neural network algorithm is used for the identification of water samples in this paper. And the morphology characteristics of the absorption spectra were used as the input parameters of water sample identification models. In the application of COD measurement, the water sample’s absorption spectra have a time correlation. Based on the foundation laid by traditional spectrum identification techniques, the algorithm was optimized in accordance with the characteristics of COD measurement. The concept of historical data queue and historical identification factor was introduced into LM-BP artificial neural network and forms the cascaded network structure. Experiments show that the method exhibited better robustness and higher accuracy than traditional algorithms, because the cascaded network is relatively more able to adapt the characteristics of the COD measurement. The test yielded a 98% identification accuracy rate, which can provide a technical guarantee for the application of spectral COD measurement in a complex environment. In this paper, the sensor structure and the proposed algorithm are simple as well, which can be used in the portable instrument with limited resource. The UV/Vis-COD measurement method based on the water sample identification algorithm can achieve improved accuracy compared to the traditional method, which calculates all water sample types with the same computational model. The proposed method is expected to solve the problem that traditional UV/Vis-COD measurement methods face regarding difficulties adapting when applied to complex environments while still achieving high COD measurement accuracy.