Multi-input-single-output (MISO) (the output variables are related to the crystal size distribution (CSD), crystal purity. and production rate, and the input variables are the fines dissolution rate, clear liquor or overflow rate, and the crystalllizer temperature) and multi-input-two-output (MITO) model predictive control of the KCI cooling crystallizer described in Part-1 of this two-part paper is investigated. The process model is the same linear ARX or the non-linear neural network models developed in Part-1. The optimization is performed by the feasible sequential quadratic programming (FSQP) algorithm (Zhou and Tits, 1992). It is shown that the non-linear MPC provides a satisfactory controller for the multivariable control of the crystallization process.