The water level of boiler drum is one of the crucial control parameters for any processrnindustries, which reflects the control of mainly boiler load and feed water indirectly. Thernincreasing demand for rapid changes in industries leads to more uncompromisingrnrequirements on the control systems for the processes. A very common control problemrnelsewhere in steam generation is that of controlling the water level in a boiler drum.rnEffectively controlling the drum water level in an industrial boiler helps to maximizernoverall energy efficiency. The problem behind various approaches used to control the drumrnwater level is lack of practical aspects such as online tuning and ease of implementationrnwithin a real plant distributed control system and also bad parameter tuning leading to poorrnlevel performance. An optimization technique can solve those problems stated byrnsimplifying the design process and allowing easily usage with physical system whilernsatisfying the optimality condition such as speed and accuracy of the response should bernwithin specified limits. rn In this thesis an optimal controller using a multivariable feedback technique using a neuralrnnetwork state estimator based linear-quadratic regulator control is introduced. The purposernof NN based LQR is to keep the level of the steam boiler drum water at the specified zerornreference value which avoids damage of boiler occurs by either overflow of water on therntop of drum which affects steam quality or by shortage of water in the drum causes therndrum burnt. Controller is designed and simulated on MATLAB/SIMULINK environmentrnand the designed optimal controller is compared with conventional PID controller.rnSimulation result indicates that PID controller reaches its steady state value of pressure ofrn5.45 bar and water level oscillates about the set with slightly increasing amplitude. The NNrnbased LQR controller achieves steady state pressure and level value of 5.455 bar andrn0.00322 mm respectively. Comparing the two controllers, the proposed neural networkrnbased linear quadratic regulator achieved good performance in both steady state responserntracking and speed of response.