The diffuse attenuation coefficient of downwelling irradiance Kd(490) is of great value for evaluating the coastal water quality and estimating the optical depth accurately. In this paper, the performance of MODIS-Aqua Kd(490) products derived using the algorithms of Muller, Werdell, Morel, Lee, and KD2M with MODIS-Aqua remote sensing reflectance Rrs(λ) and Chl-a products were compared in the South China Sea (SCS) based on the relaxed match-ups between MODIS-Aqua products and in situ data collected from 2004 to 2012. Further, taking into account the unique optical properties of the SCS waters, the performance of Kd(490) products derived using the above algorithms with field reflectance Rrs(λ) and Chl-a data were evaluated in order to analyze the causes of the uncertainties of Kd(490) products and the necessity of regional algorithm optimization. The results show that MODIS-Aqua Kd(490) products derived using the algorithms of Muller, Werdell, and Morel with MODIS-Aqua Rrs(λ) and Chl-a products underestimated the in situ data, while the algorithms of Lee and KD2M overestimated the data, with average relative errors (APD) of 20.12%, 29.22%, 17.98%, 22.61%, and 21.61%, respectively. The algorithm of Morel performed the best, with a coefficient of determination R2 of 0.5. There are significant improvements in the accuracy of the above algorithms by using field Rrs(λ) and Chl-a data, especially for the algorithms of Werdell and Lee, with an APD of 5%—11% improvement. The algorithm of Morel performed the best, with an APD of 15.44%. It is necessary to choose the proper algorithm and conduct algorithm regionalization correction for the accuracy estimation of Kd(490) products in the SCS.