In this paper, we propose a new approach to perform cell dielectric characterization from their electrorotation spectrum. At first, a variance analysis is carried out to quantify the dispersion in electrorotation spectra due to the different parameters involved. On this basis, the impact of each parameter is emphasized by weighing the spectrum with an appropriate frequency-dependent coefficient: this technique enables to minimize the coupling effects which deteriorate the accuracy of parameter extraction. In addition, the Nelder-Mead simplex algorithm used in the identification procedure is modified to account for bounded intervals in which the unknown parameters are expected to vary. Both these techniques have proven to give increased confidence levels compared to previous work reported in the literature.