Ultraviolet/visible (UV/Vis) spectroscopy was investigated for the rapid determination of chemical oxygen demand (COD) which was an indicator to measure the concentration of organic matter in aquaculture water. A total number of 135 collected turtle breeding water samples were scanned for UV/Vis spectrum, uninformative variable elimination (UVE) and successive projections algorithm (SPA) were combined as a mixed variable selection method to perform characteristic wavelength selection from the full wavelength spectrum, 7 characteristic wavelengths were selected from full 201 UV/Vis spectral variables, which were just 3.48% number of the full range spectrum, and the calibration time and complexity of the modeling were greatly reduced. The predicted results which were obtained by using least squares-support vector machine (LS-SVM) calibration showed that the characteristic wavelengths achieved better results (0.89 for correlation coefficient (r), 15.46 mg·L-1 for root mean square error of prediction (RMSEP)) than full wavelengths did (0.88 for r and 15.71 mg·L-1 for RMSEP). The comprehensive results revealed that the UV/Vis characteristic wavelengths which were obtained by UVE-SPA variable selection method, combined with LS-SVM calibration could apply to the rapid and accurate determination of COD in aquaculture water. Moreover, this study laid the foundation for further implementation of online analysis of aquaculture water and rapid determination of other water quality parameters.