A fluorescence spectroscopy method for differentiation of phytoplankton classes is developed. SA4 multiwavelet functions are used to decompose the three-dimensional (3D) fluorescence spectra date of 42 phytoplankton species dominant in coastal area of China sea into scale vectors and wavelet vectors, then the Ca2, Ca3 and Ca2-Ca3 vectors are chosen as fluorescence feature spectra by Bayesian discriminant analysis and the three kinds of reference spectra are obtained by application of the systematic cluster to the fluorescence feature spectra, based on which, a fluorescence spectroscopy method is developed by multivariate linear regression resolved by nonnegative least squares. The three kinds of reference spectra are utilized to the samples composed of one phytoplankton species, the results suggest that the Ca2 reference spectra are the strongest: the average correctly discriminating rate are 96.1% at the division level and 87.4% at the genus level. For the simulative mixed samples, the correctly discriminating rate for dominant species ranged from 91.4% to 100%, the average are 98.0%, 99.4%, 99.9%, the average relative content are 72.7%, 77.1%, 86.1% at division level, the correctly discriminating rate for dominant species ranged from 23.2% to 100% and the average are 91.2%, 92.5%, 93.4% at genus level in three different ratios respectively. For the mixed samples in lab, the correctly discriminating rate for dominate species range from 60.0% to 100%, the average is 97.0%, and the average relative content is 71.8% at division level; the correctly discriminating rate for dominate species range from 25.0% to 100%, the average is 83.1% at genus level. For the samples from the mesocosm experiment in Maidao Bay of Qingdao and Jiaozhou Bay, the discriminating results of dominant phytoplankton classes from the fluorescence spectroscopy method are the same as that of microscopic examination at division level, and for the 10 samples with over 75% relative dominance of one phytoplankton species, the dominant species of 7 samples are recognized at genus level.