The detection of parameters for water quality with spectral technique is a research hotspot at present. This paper proposes a method for the determination of chemical oxygen demand (COD) based on the fluorescence emission spectrum. Two groups of experimental samples are provided in the experiment, among which the fundamental group is 20 COD standard solutions, and the remaining 63 are actual water samples as the other group. Rapid digestion spectrophotometry is utilized to detect the COD of experimental samples. Three dimensional fluorescence spectrophotometer is used to collect the fluorescence emission spectra of the water samples at EX=275 nm (all the range of fluorescence emission spectra are EM=325~450 nm), then the data of fluorescence emission spectra of two kinds of water samples are processed and modeled. Principal component regression (PCR) and partial least squares regression (PLSR) are utilized to establish the prediction models based on fluorescence emission data respectively, and the effects of the models are compared. In order to verify the feasibility of the proposed method and the prediction ability of the model, the results of the PLSR mode are compared with the standard method. The comparison results show that, for the COD standard solution, when the number of principal component of PLSR and PCR is 5 and 8 respectively, the optimal results are obtained for both models, of which the determination coefficients of the correction model are R2PLS=0.999 9 and R2PCR=0.989 7, respectively. The prediction error of validation set data in the calibration model is less than 10%, and the PLSR model is better than the PCR model. While for the actual water samples, when the number of principal component of PLSR and PCR is 6 and 7 respectively, the cross-validation effect of the correction model is the best. Among them, root mean square error of cross-validation of the PLSR method is RMSECVPLS=0.932 2 mg·L-1, while for the PCR algorithm, RMSECVPCR=0.976 4 mg·L-1. For the validation set, the determination coefficient of PLSR is 0.940 2, while for PCR method, it is 0.919 0. It shows that PLSR method has better prediction effect. Consequently, the PLSR model based on fluorescence emission spectrum data has high prediction ability and strong adaptability, which can detect water COD quickly and accurately. Through the comparison of the method proposed in this paper and the traditional detection, we can see that proposed method can be used to detect the water with low concentration of organic pollutants, however, the detection error will increase if the concentration of organic pollutants in the detected water is high. This paper providesa new design idea for the research and development of water quality detection optical sensor.