Control structure selection (CSS) deals with the problem of selecting appropriate manipulated and controlled variables and is the very first and crucial step when designing any control system. The limitations on the system's performance and the dynamics of the controlled system are largely determined by the underlying control structure. However, little attention was paid to CSS in the literature from the control community and the focus of most of the work is on the tracking performance rather on the economic performance of the resulting controlled system.The main contribution of this thesis is to extend the previous work of Tobias Scharf which is based on the evaluation of economic performance of control structures for stationary process to include the dynamic performance in a consistent manner. The basic idea is that a feedback control system when regulating a set of controlled variables to their set-points should steer the process toward the economic optimum in the presence of disturbances and plant model mismatch. Nonlinear model predictive control (NMPC) is assumed to avoid the problem of the dependence of the dynamic performance on the type and the parameters of the controller. The weights of the NMPC are optimized based on the resulting profit of the system. Realistic dynamic disturbances are considered, not only the worst-case disturbances are taken into account but also the probability of the occurrence of smaller disturbances which occur more frequently. Thus the result is a good approximation of the attainable dynamic performance of each structure in reality and the structures are compared on an equal ground.A systematic control structure selection procedure is proposed which consists of 6 steps. The first step is to define the process degrees of freedom, the economic performance to optimized, the requirements to be fulfilled and the candidate manipulated/controlled variables. The disturbance scenarios that are expected to occur are defined in the next step. As the number of possible structures is often very large, a pre-screening step is needed to reduce it. In step 4, the set-points for regulatory control are optimized with all disturbance scenarios being taken into account. In step 5, the static worst-case performance of each control structure is investigated to see whether it can guarantee an optimal process operation in the presence of disturbances. The dynamic control performance is evaluated in the last step.The methodology is applied to two case studies, a ternary distillation column that separates a mixture of methanol, ethanol and 1-propanol, and a reactive distillation column that produces methyl acetate from acetic acid and methanol. The results show that there is a strong correlation between the controlled variables chosen by our procedure and the result from the so-called generalized non-square relative gain array (NRGA) which is a generalized version of the very well-known relative gain array (RGA). The theoretical analysis at the end of this thesis verifies that the NRGA value of a controlled variable indeed implies a lower bound on the loss of the economic performance and thus, the variables with larger NRGA values tend to achieve better performance.