We report the application of the Model-based Predictive Control (MPC) to improve the performance of the start-up of a 150-175 MW combined cycle power plant whose gas turbine is fueled by natural gas. In concrete the simulations have shown that the efficient drum level control is reflected on the improvement of power efficiency in the sense of reaching the 225 MW set point in around 45 minutes faster than the case of PID. Experimental data taken from ordinary runs from power plant was used for ends of system identification which is based on convolution integrals resulting well adjustable to the acquired data. Simulations have demonstrated that the performance of the MPC surpasses to the one of classic PID essentially in two aspects: (i) reducing the time for reaching set point and (ii) avoiding unexpected critical situations during the plant start-up. Results have indicated that the MPC might reduce in up to 45±5 minutes the time of reaching the set point established to be 225MWwithin a computational error of 5%, which is translated as the MPC error of order of 2.5% working as software in plant. All these results might sustain the fact that the MPC based on convolution models appears to be an interesting scheme to optimize the full functionality in power plants whose expected power is ranging between 200 and 250 MW.