This work discusses the maintenance of the water level for steam generators at its programmed value. The control algorithm used here incorporates pole-assignment into minimum variance strategy. Therefore, the control system can keep the water level constant during full power by locating closed-loop poles appropriately, although the process has the characteristics of high complexity and nonlinearity. For computer simulation, the parameter estimation is based on a parallel adaptation algorithm that is robust to noises. Increased computation efforts can compensate for lack of an integrator that is required to eliminate a steady state error. The steady state error is cancelled by a constant gain. All parameters are updated on-line at every sample instant, and control actions occur at every M sample instant in order to exclude excessive control actions. Also, the extra perturbation signals are added to the input signal such that the control system guarantee persistently exciting. As shown in the results, excessive actions are diminished under low power conditions, too. And the stability of the closed-loop system is ensured at full power, because the closed-loop poles are placed to stable locations. Thus serious fluctuations in the water level hardly occur at full power, although the system is highly complex, nonlinear, and time-varying. As a result, the present poleassignment self-tuning controller is quite robust and can easily deal with the unstable system or the system that has nonminimum phase property.