Chemical oxygen demand (COD) is an important indicator of organic matter pollution in water source and one of the necessary parameters for monitoring water quality. Detecting the COD based on ultraviolet spectrophotometry is an important method for online real-time monitoring of pollution sources. This study designs a COD online detection sensor based on a dual-light source to solve the problems of COD online detection using the traditional reagent method. The influences of turbidity, light-source temperature, and water-sample temperature on measurement are analyzed theoretically. Methods of temperature and turbidity compensations are then proposed. Experimental results show that the repeatability error of the sensor is 3.12%, whereas the maximum relative error of the actual water-sample comparison experiment is 6.18%. The turbidity and temperature compensations are accurate. The designed COD sensor can accurately measure the COD concentration of the water samples to meet the technical requirements of water quality online monitoring equipment.