The research on characteristic wavelengths analysis of reflectance spectrum is a very important and basic task for remote sensing of inland-water color. The present paper analyzed remote sensing reflectances of 312 samples measured in Taihu Lake between 2006 and 2009, and these reflectances were separated into three classes by chlorophyll-a concentrations. The reflectance spectra smoothed by Savitzky-Golay algorithm were calculated by first- and second-order derivatives. Then, zero values were located in the derivatives and counted at all wavelengths. Thus the frequency distribution of zeros at each wavelength was got. At which wavelength a local maximum of the frequencies appears a characteristic wavelength will most likely be there. These characteristic wavelengths are corresponding to maximum, minimum, from-concave-to-convex inflection point and from-convex-to-concave inflection point of a spectrum curve. At last the paper provided the characteristic wavelengths for Taihu Lake water at the spectral coverage from 350 to 900 nm, which are 359, 440, 464, 472, 552, 566, 583, 628, 636, 645, 660, 676, 689, 706, 728, 791, 806, and 825 nm. In addition, these wavelengths we found were explained by absorption of phytoplankton pigments and components of water in Taihu Lake. Being able to distinguish overlaps between peaks and vales at the same wavelength in different measurements, the method to analyze characteristic wavelengths is universally applicable to various spectrum curves. The characteristic wavelengths chosen by the paper are helpful to improving some algorithms of retrieval of water quality parameters.