Based on a newly developed mathematical model. the complex dynamic simulator of an industrial Universal Oil Products (UOP) fluid catalytic cracking unit was used to implement the model predictive control (MPC) algorithm, The simulator revealed the multivariable, nonlinear and strong interacting feature of the process. Combined with equipment and operating constraints they put severe limits on control performance. Different MPC schemes for the reactor and regenerator's most important process variables were tested and the most favorable have been presented. The constrained MPC approach using scheduled linearization to account for non-linear behavior and a larger number of manipulated than controlled variables proved successful. Comparison with traditional control using decentralized PID controllers revealed incentives for the multivariable model based predictive control in maintaining controlled variables very close to their constrained limits where usually the optimum is situated.