On the basis of the survey data collected over eight cruises from 2006 to 2015 in the South China Sea, the correlation features between the spectral absorption coefficient of phytoplankton and the typical pigment concentrations in the South China Sea were investigated by the derivative spectral analysis method, and partial least squares (PLS) regression models were then developed for estimating the pigment concentrations. The study results show that the derivative spectral variation of phytoplankton absorption coefficients can be used for quantitatively estimating the key pigment concentrations, and the estimation results based on the second-order derivative spectra are slightly better than those based on the fourth-order spectra. The proposed models achieve high accuracy for TChl a, PSC, Fuco, 19But, 19Hex, and Diadino, with a high linear correlation between the predicted and measured values. However, the estimation accuracy of PPC and Zea is relatively low. Compared with the empirical models based on the TChl a concentration, the PLS regression models based on derivative spectra have an equivalent effect on estimating the pigments such as PSC, Fuco, 19Hex, and Diadino, and show certain advantages for 19But, PPC, and Zea. The internal relationship between the concentration of different types of pigments and the derivative spectra of absorption coefficients may be an important factor for the difference in the estimation accuracy of different pigments. In conclusion, the method of quantitatively extracting the pigment concentration by the derivative spectra based on the phytoplankton absorption coefficients established in this paper can provide a reference for deeply investigating the hyperspectral remote sensing of phytoplankton population structures based on the ocean color satellites.