Following the recent successes with low input data variability and soft sensor design under feed composition changes, this study proposes, among other things, the use of a physics-based approach to improve multivariable model predictive control (MPC) of naphtha distillation. Unlike the industrial settings, where the influence of other manipulated variables is difficult to exclude due to the actions of the human operator, a physically based modeling provides close to an ideal step-by-step and one-by-one testing of the chemical process, resulting in improved accuracy of the transfer function matrix used for MPC design. The proposed approach has been tested on canonical and alternative control schemes used in stabilized naphtha production. Importantly, the physics-based model resolved all the issues associated with unavailability to reach the set points in controlling the quality of end products when compared with MPC built on the industrial data only irrespective of the control scheme considered. As a result, the steady-state controllability analysis and the closed-loop process behavior highlight that an alternative control structure with transfer function matrix obtained on a physics-based model is a better choice for the industrial case study. Thus, the developed strategy for MPC design was approved as relevant for the cases when a preliminary control scheme requires an update or optimized control scheme without affecting production is of great demand.