This paper deals with the design and on-line implementation of a real-time dynamic optimizer for fed-batch evaporative crystallization of an ammonium sulphate-water system. An optimal heat input profile to a 75 liter draft tube crystallizer is sought such that an objective function pertinent to the batch crystal yield is maximized while the reproducibility of batches and the fulfillment of desired product specifications are sustained. The seeded fed-batch crystallizer is represented by a non-linear moment model that is experimentally verified to be an adequate description of the process at hand. The optimal control problem is solved by the sequential approach. Due to lack of supersaturation measurements, the evolution of supersaturation during the process is estimated using the extended Luenberger observation technique. A closed-loop implementation of the dynamic optimizer reveals that a substantial increase in the crystal volume fraction at the batch end can be attained by an optimal heat input profile, whereas the impact of the heat input on the median crystal size is insignificant.