Phytoplankton absorption coefficient is an important parameter for ocean optics. The parameter varies widely with different locations and time. In order to model the phytoplankton absorption coefficient, application of the cluster analysis on the retrieval of phytoplankton absorption coefficient based on a non-linear model of phytoplankton absorption coefficient is discussed. The method reduces the error due to different optical properties of phytoplankton in different areas. By calculating adjusted Rand index (ARI) between the second derivative reflectance and absorption spectra at different wavelengths, the spectral range of 555~681 nm is found to be more sensitive than other bands. The spectral band of 555~681 nm is applied to analyze the second derivative reflectance. In situ data are used to validate the cluster analysis method, and it is found that the cluster analysis is effective for phytoplankton absorption model (root mean square error is higher than 0.79). The study is a fundamental research for application of optical remote sensing to retrieving phytoplankton information and ocean component.