Understanding and controlling aggregation is an essential aspect in the development of pharmaceutical proteins to improve product yield, potency and quality consistency. Even a minute quantity of aggregates may be reactogenic and can render the final product unusable. Self-assembly processing of virus-like particles (VLPs) is an efficient method to quicken the delivery of safe and efficacious vaccines to the market at low cost. VLP production, as with the manufacture of many biotherapeutics, is susceptible to aggregation, which may be minimized through the use of accurate and practical mathematical models. However, existing models for virus assembly are idealized, and do not predict the non-native aggregation behavior of self-assembling viral subunits in a tractable nor useful way. Here we present a mechanistic mathematical model describing VLP self-assembly that accounts for partitioning of reactive subunits between the correct and aggregation pathways. Our results show that unproductive aggregation causes up to 38% product loss by competing favorably with the productive nucleation of self-assembling subunits, therefore limiting the availability of nuclei for subsequent capsid growth. The protein subunit aggregation reaction exhibits an apparent second-order concentration dependence, suggesting a dimerization-controlled agglomeration pathway. Despite the plethora of possible assembly intermediates and aggregation pathways, protein aggregation behavior may be predicted by a relatively simple yet realistic model. More importantly, we have shown that our bioengineering model is amenable to different reactor formats, thus opening the way to rational scale-up strategies for products that comprise biomolecular assemblies. Biotechnol. Bioeng. 2010;107: 550–560. © 2010 Wiley Periodicals, Inc.