This work analyzes the effect of the process design on control structure for a system with a reactor, two distillation columns, and two recycle streams. The reaction A + B --> C occurs in a reactor, and since component C is assumed to be the intermediate boiler, the two distillation columns recycle components A and B back to the reactor. A previous paper presented two workable control structures for this process. One fixed the flow rates of the two recycle streams and brought in makeup fresh feeds of components A and B on level control. The other control structure fixed the reactor effluent flow rate, controlled the composition of one reactant in the reactor by manipulating one fresh feed, and brought in the other fresh feed on reactor level control. These two structures have the undesirable feature of not being able to set directly the production rate and, in the second structure, requiring a reactor composition measurement, which can be difficult due to the hostile environment and can require expensive instrument maintenance. Studies of other more complex processes have led to similar results : measurement of composition somewhere in the reaction section is necessary for stable operation. In this paper we present an analysis that explains the fundamental problem with control structures in which one fresh feed is fixed and no reactor composition is measured. We show that this control structure can work if modifications are made in the design from the steady-state economic optimum. This highlights the potential trade-off between steady-state economics and dynamic controllability and illustrates considerations that ought to be included during the conceptual design procedure. A modified control structure is proposed that provides effective control of the economically optimal process design. It permits throughput to be directly set and does not require a composition measurement. The basic idea is to use the flow rates of the recycles from the separation section to infer reactor compositions. Dynamic simulation studies on both simplified and rigorous models are used to evaluate the performance of the proposed control system over a wide range of reactor sizes.