The fluid catalytic cracking unit (FCCU) has a major effect on profitability of an oil refinery. The FCCU is difficult to model well, due to significant nonlinearities and interactions. Control of the FCC is challenging and there is strong incentive to use multivariable (MV) control schemes, such as model predictive control (MPC), which accommodate these interactions. The linear MV schemes rely on linearized model around an operating point and therefore, it is difficult to obtain high quality control. This paper uses a singular value decomposition method (N4SID) to obtain a state space model which is then reduced to a step model required by the MPC algorithin. Simulations on a detailed model [Chem. Eng. Comm. 146 (1996) 163; Trans. IchemE 75 (1997) 401; A modified integrated dynamic model of a riser type FCC unit, Master's Thesis, University of Saskatchewan, Saskatoon (1998); Can. J. Chem. Eng. 77 (1999) 169] show the effectiveness of this approach.