Polarized light backscattering spectra invertion model is established for epitheliumlike tissue phantom based on particles model of biological tissue and sensitivity of spectra to Mie scatters morphology. This inversion model is of multiple parameters, multiple extreme values and nonlinear, complicated trigonometric calculation is involved, and the model space is limited. Thus, floating genetic algorithms ( FGAs), which reserves the optimal unit for every epoch, are applied to invert polarized light backscattering spectra for epitheliumlike tissue phantom. The strategies for encode, fitness adjusting and selection are discussed. The results indicate that, with 70 iteration epochs the relative error for every parameter inclines to be stable, about 0.02% for the minimal and less than 3% for the maximal. The genetic algorithm based on real encode can obtain the morphological parameters of the upper layer simultaneously by inverting the polarized light backscattering spectra, with global convergence, high inversion precision and robustness.