ABSTRACTRecent advances in multi-region input-output (IO) table construction have led to large databases becoming available. Some of these databases currently demand too much computer memory or user cognition to be handled effectively outside high-performance environments, especially for applications such as virtual laboratories, computable general equilibrium modelling, linear programming, series expansion, or structural decomposition analysis, thus inhibiting their widespread use by analysts and decision-makers. Aggregation is an obvious solution; but there is a need for structured approaches to aggregating an IO system in a way that does not compromise the ability to effectively answer the research question at hand. In this article, I describe how structural path analysis can be used to realise a computationally inexpensive method for aggregating IO systems whilst minimising aggregation errors. I show that there exists no one-fits-all strategy, but that optimal aggregation depends on the research question at hand.