Nonlinear model predictive control algorithm is used for the on-line control of polymer molecular weight distribution. The control of chain-length distribution is achieved by selecting a collection of points in the distribution and using it as set points for the control algorithm. An on-line Kalman filter is used to incorporate infrequent and delayed off-line molecular weight measurements. Through simulation; the control algorithm is evaluated, under tracking conditions as well as plant-model mismatch. The results demonstrate that the control algorithm can regulate the entire molecular weight distribution with high computational efficiency and minimum steady state error