Dynamic Matrix Control, a predictive computer control algorithm developed at Shell Oil Co. (Cutler, 1983) is applied to dual composition control of moderate and high purity, binary distillation towers. The composition control loops in these towers are highly interacting and, in addition, very nonlinear. The difficulty in obtaining a representative process model for nonlinear systems is demonstrated and methods to overcome this problem are suggested. Multivariable gain and time constant scheduling are evaluated using simple analytical models to update key process model parameters (process gains and time constants) on-line. This approach results in a significant improvement in regulatory control performance over the standard application of DMC.