High-density cultures of the methylotrophic yeast Pichia pastoris were found to exhibit oscillatory metabolic behavior when fed methanol under closed-loop operations using a dissolved oxygen-based bioreactor feed controller (DOstat). This behavior, if left unattended, led to the irreversible loss of culture productivity, 1 to 2 days after growth on methanol commenced, presumably through the accumulation of incompletely oxidized intermediates. To provide insights into how fermentation operation conditions and strain variations might contribute to this phenomenon a theoretical study was initiated. In this article, a simple mathematical model of the closed-loop DOstat is developed and analyzed with the goat of deriving theoretical stability criteria applicable to the design of metabolic feed controllers during high-cell-density fermentations. The model consists of a system of differential, integral, and algebraic equations describing the biological process and the components of the standard proportional integral (PI) feedback controller. Inputs into the process model include metabolic pathway information, oxidative metabolism stoichiometry, and substrate uptake kinetics. Frequency-response analysis and the Bode stability criterion are applied to derive controller stability criteria with particular emphasis on elucidating the role(s) that model parameters, both biological and operational, have on system stability. The results of this analysis are used to construct a closed-loop fermentation-operating diagram relating fermentation operating parameters to the oxidative capacity of the culture. The utility of this analysis is demonstrated through the application of the results to the design and stabilization of the DOstat during high-density fermentations of P. pastoris growing on methanol or glycerol. From this analysis, it is possible to conclude that, when the rate of oxygen transfer approaches in magnitude the rate of oxygen utilization, the potential for controller destabilization is greatest. Under these conditions, the region of parameter space associated with stable controller operations is further constrained. Because the only system-specific experimental inputs into the model are the routinely measured residual substrate and dissolved oxygen concentrations, the framework presented should provide simple and practical theoretical guidelines relevant to the design of similar industrial fermentation feed controllers independent of the specifics of the biological system at hand. (C) 2000 John Wiley & Sons, Inc.