Dissolved organic carbon (DOC) is the total carbon content of compounds dissolved in water, which is an important indicator of the marine carbon cycle and ecosystems and is of great significance to the carbon sink capacity of the oceans and the maintenance of marine ecosystems in equilibrium. The commonly used detection method for DOC in seawater is the laboratory chemical analysis method, but this detection method is complicated in operation and cumbersome in processing, and it does not have the ability to detect all-weather rapidly. To this end, this paper presents a method for detecting DOC in seawater based on ultraviolet absorption spectroscopy. A prediction model of DOC in seawater was constructed by partial least squares (PLS) and calibrated by multiple linear regression (MLR) methods, which completed the detection of DOC in 10 sets of actual unfiltered seawater samples. The results showed that using the constructed model to correct the spectral data containing impurities such as particles, significantly enhanced the correlation of measured dissolved organic carbon. Specifically, the linear correlation coefficient R2 improved from 0.8891 to 0.9838. Furthermore, the mean absolute error (MAE) decreased from 9.775% to 3.781%, and the individual absolute errors were maintained within 9%. The PLS-MLR model achieves accurate measurement of DOC in seawater based on ultraviolet absorption spectrometry without pre-processing, and the method provides a theoretical basis for rapid detection and analysis of DOC in water bodies and the development of sensors.