This paper presents several comparative results for three output feedback controllers, i.e., proportional-integral (PI) control, state-space formulation of nonlinear internal model control (NIMC), and dynamic output feedback control (DOFC), for exothermic parallel reactions in a continuously stirred tank reactor. Under the known measurement delay, initial operating points, and multiplicity behavior of the process, the linear PI controller with the Z-N tuning method can provide the no offset performance for set point change and disturbance rejection. Since the NIMC design incorporates an open-loop model, the output regulation of an unstable system may be obviously degraded within large operating region. The DOFC design which combines a dynamic state feedback through Fliess＇ linearization algorithm with a nonlinear Luenberger-type observer can provide a smooth and stable regulation response in an arbitrary operating region. With the aid of a complete theoretical analysis, it is also found that the DOFC design is robust to uncertainty and measurement delay.