The problem of scheduling multiple feeds on parallel units is addressed The performance of each unit is assumed to decrease with time, therefore requiring planned maintenance to restore its performance. A mixed-integer nonlinear programming (MINLP) model presented uses the process information regarding the exponential decay in performance with time to find a cyclic schedule for feed processing. Optimization trade-off is between the performance of the units, and the maintenance costs and the loss in production due to shutdowns. Three main decisions for the optimal schedule are : a) assignment of various feeds to different units; b) processing time of each feed on different units; and c) number of subcycles of each feed on different units. Special mathematical properties of the MINLP model are exploited to solve the problem to global optimality with a branch-and-bound method. Using several examples advantages of the model are illustrated and performances of different solution algorithms are compared.