Polarized light scattering spectroscopy (LSS) is sensitive to the cell nuclear morphological changes in the various forms of epithelial dysplasia. Extensive studies illustrate it is a promising in situ technique to detect precancerous and early cancerous changes in the epithelial tissue. To determine the density and size distribution of cell nuclei with spectra, generally, Mie theory-based inverse model is adopted. This model is of multiple parameters, multiple extreme values and nonlinear. The determination of all unknown parameters needs to solve a nonlinear inverse problem. Other than least-square fitting used by previous studies, in this paper, we developed a novel method - float genetic algorithm (FGA) to determine the particle size distribution and refractive index for LSS. Our results showed that, relative errors of three estimated statistical quantities: diameter, standard deviation and refractive index are less than 5% for different additive Gauss noise levels with 70 iteration epochs. The errors gradually decrease with iteration epoch increases. Moreover, comparing with Newton-type iteration method coupled with a Marquardt-Tikhanov regularization scheme, FGA avoids the problems of local extreme value and selection of initial value and regularization parameters, thus obtains the advantages of high precision, stability and robustness.