Direct chemiluminescence emission from the reaction of acidic permanganate and organic compounds was employed for determining the chemical oxygen demand (COD) in water (1-step CL COD). Due to the diversity of organic pollutants in water, there are no standards for COD measurements, and many compounds do not show any chemiluminescence signal in the 1-step CL COD method. As a result, this method shows a low correlation with the conventional CODMn method. In this study, a new 3-step CL COD method was developed to overcome these drawbacks. The basic principle of the 3-step CL COD method is based on the principle of "back titration" in the CODMn method: (i) the sample is treated with permanganate under heating, (ii) the excess permanganate is treated with pyrogallol, and (iii) the excess pyrogallol is measured by the chemiluminescence reaction with permanganate. The reagent concentration, sample volume, and heating temperature were optimized, and the 3-step CL COD method successfully obtained the signal from some samples that cannot be detected by 1-step CL COD method. The calibration graph is linear in the range of 0 -12.86 mg/L with a detection limit of 0.082 mg/L. This method is continuous, sensitive and low cost compared with the conventional method, and is applicable for on-site monitoring. The effect of the chloride ion was investigated, and showed an insignificant effect after two-times dilution of high-salinity samples. The correlation with the CODMn method for various organic compounds showed a good coefficient of determination, R 2 = 0.9773 (n = 16).