The complexity of biological systems makes physiological bioprocess control a challenging task. The establishment of physiological process control systems using model based control is key in the development of advanced bioprocess control systems. Advanced bioprocess control strategies, which aim at avoiding unwanted by-products rather then removing them, are recommended by regulatory authorities such as the US Food and Drugs Administration (FDA). In order to do so, models which are able to accurately predict the behaviour of the cells during their cultivation are needed. However, developing such models is a difficult task, since the production of recombinant products causes a lot of cellular stress for the used production hosts, which negatively effects the performance of the cells, resulting in a decline in specific performance. The accurate prediction of the growth and productivity of the cells is necessary in order to set up and facilitate an optimal control strategy for each individual bioprocesses. For this study two different mechanistic modelling approaches to model the performance of Escherichia coli, which allow for model predictive control (MPC), are compared to each other. Thereby, the quality of fit as well as the sensitivity and identifiability of the models is analysed. The first approach, modelled the performance decay of the cells during the cultivation via the cumulative metabolised substrate, and relied only on the gas rates as monitored states of the cultivation. For the second approach, additionally the cell size was monitored as characteristic for the physiological state of the cells used during the process. Hereby, the specific product formation rate of the cells was used to model the performance decay of the cells, instead of using the consumed amount of sugar. The model for the second modelling approach, the cell size model, was newly developed for this study. Both models were analysed and compared towards each other in terms of model fit (using NMRSE values as well as their standard deviation as quantitative measure and time-resolved as well as observed-vspredictied plots as qualitative measure) and model structure (using local sensitivity analysis and structural identifiability analysis). With the newly developed model a more accurate description of the system with lower NMRSE and standard deviation of NMRSE values (< 11 % NMRSE instead of < 20 % and < 9 % StDev of NMRSE instead of < 15 %) as well as the increased structural identifiability of its parameters (all parameters at once instead of 2 sets of parameters) could be achieved. i It could be shown that both used modelling approaches are giving a valid description of the system. Furthermore, by using the additional monitored state of the cell size, the prediction accuracy and reproducibility could be enhanced. Hereby, more accurate predictions by mechanistic models of industrial relevant bioprocesses, due to an increased insight in cell physiological reactions, can be used to decrease the formation of by-products, to more reliably reach the needed product quality as well as to increase the overall space-time-yields of production plants. However, to apply the newly developed cell size model for industrial bioprocesses, further testing of the model with predictive control needs to be done, in order to compare its control capabilities to commonly applied standard control strategies.