The present research was targeted to develop a fluorescence analyser for phytoplankton population which uses a series of LEDs as the light source. So the 3D discrete fluorescence spectra with 12 excitation wavelengths (400, 430, 450, 460, 470, 490, 500, 510, 525, 550, 570 and 590 nm) were determined by fluorescence spectrophotometer for 43 phytoplankton species. Then, the wavelet, Daubechies-7 (Db7), and Bayes Classifier were applied to extract the characteristics for each classes from the 3D discrete fluorescence spectra. Lastly, the fluorescence differentiation method for phytoplankton populations was established by multivariate linear regression and non-negative least squares, which could differentiate phytoplankton populations at the levels of both divisions and genus. This method was tested: for simulatively mixed samples(the dominant species accounted for 70%, 80%, 90% and 100% of the gross biomass, respectively) from 32 red tide algal species, and the correct discrimination rates at the level of genus were 67.5%, 75.8%, 81.4% and 79.4%, respectively. For simulatively mixed samples (the dominant divisions algae accounted for 50%, 75% and 100% of the gross biomass, respectively) from 43 algal species, the discrimination rates at the level of division were 95.2%, 99.7% and 91.9% with average relative content of 38.1%, 63.2% and 90.5%, respectively.