The competing first- and third-order reaction scheme for lysozyme is shown to not predict fed-batch lysozyme refolding when the model is parameterized using independent batch experiments, even when variations in chemical composition during the fed-batch experiment are accounted for. A new kinetic scheme is proposed that involves rapid partitioning between the alternative fates of refolding and aggregation, and which allows for aggregation via a sequential mechanism. The model assumes that monomeric lysozyme in different states, including native, is able to aggregate with intermediates, accounting for recent experimental evidence that native protein can be incorporated into aggregates and explaining why native protein in the refolding buffer reduces yield. Stopped-flow light-scattering measurements were used to measure the association rate for the sequential aggregation mechanism, and refolding rate constants were determined in a series of batch experiments designed to be "snapshots" of the composition during a fed-batch experiment. The new kinetic scheme gave a good a priori prediction of fed-batch refolding performance. (C) 2003 Wiley Periodicals, Inc.