AbstractA model was formulated to examine the competitive growth of two phenotypes (Leu+and Leu−) and the product formation with recombinantSaccharomyces cerevisiaestrain DBY‐745, which contains the shuttle vector pYGH3‐16‐s with the foreign gene HBsAg (hepatitis B virus surface antigen) as well as experimental fedbatch fermentation data. The important state variables and the process parameters evaluated include (1) the ratio of the plasmid‐free cell concentration to the plasmid‐containing cell concentration (ρ =X−X+), (2) the expression of human hepatitis B surface antigeng(CH), (3) the glucose consumption (S), (4) the ethanol production (/), (5) the change of working volume (V) in the fermentor, (6) the different specific growth rates of two phenotype cells, and (7) the plasmid loss frequency coefficient (α ). These variables and other parameters were carefully defined, their correlations were studied, and a mathematical model using a set of nonlinear ordinary differential equations (ODEs) for fed‐batch fermentation was then obtained based on the theoretical considerations and the experimental results. The extended Kalman filter (EKF) methods was applied for the best estimate of these variables based on the experimentally observable variables: ρV, andg(CH). Each of these variable was affected by random measuring errors under the different operating conditions. Simulation results presented for verification of the model agreed with our observations and provided useful information relevant to the operation and the control of the fedbatch recombinant yeast fermentation. The method of predicting an optimal profile of the cell growth was also demonstrated under the different dissolved oxygen concentrations. © 1993 J