For spectral information redundancy and correlation in phytoplankton spectral analysis, interval Monte Carlo partial least squares (IMC-PLS) which effectively solves the problem of feature wavelength selection is presented based on partial least squares (PLS). Feature region is preselected according to the location of the pigment fluorescence peaks, the internal informations of a single band and the contributions of different random band combinations to the model are plenarily used. Based on three-linear feature of fluorescence spectra, emission wavelength band and excitation wavelength band are considered as a unit. The result shows that comparing with the uninformative variable eliminotion (UVE), feature wavelength points and computation time obtained by IMC-PLS decrease by 80.1% and 81.3% and average relative tolerances (ARTs) by inversion of four algae concentrations decrease by 0%, 34.3%, 55.9%, 30.5%. IMC-PLS algorithm effectively solves the problem of real-time monitoring, and provides theoretical support for the development of a discrete three-dimensional fluorescence spectrometer meanwhile.