Accurate online monitoring of biomass is an enduring problem. Of the various methods available, one of the more satisfactory is with dielec. spectroscopy (often referred to as capacitance monitoring). However, as this technique measures total -biovolume', large changes in cellular morphol. or population distribution will alter the calibration with respect to cell wt. or no. In an attempt to minimize this problem, the full dielec. spectrum was scanned every 10-20 min, and then multivariate techniques were used to build predictive models for dry cell wt., cell no. and median cell diam. The method was applied to high cell d. baker's yeast fed-batch and CHO perfusion culture. Exptl. results clearly showed the enhanced value of capacitance monitoring over range of frequencies. For example, a phase plot of the capacitance at a low excitation frequency vs. the value at a higher frequency was an accurate indicator of the major transition points of CHO perfusion culture, i.e. max. cell viability, end of lactate consumption, point of zero viability. [on SciFinder (R)]