Composite faecal egg counts (FEC) are increasingly used to support strategic anthelmintic treatment decisions in grazing livestock. However, their accuracy as estimators of group mean FEC is affected by the number of individual samples included, how thoroughly they are mixed, and the underlying degree of parasite aggregation between individual hosts. This paper uses a Negative Binomial model for parasite aggregation, and a Poisson model for egg distribution within faecal suspensions, in order to optimise composite FEC protocol for commercial sheep flocks. Our results suggest that faecal egg density in a well-mixed composite sample from 10 sheep (3 g of faeces from each), estimated by examination of four independently filled McMaster chambers, is likely to provide an adequate estimate of group mean FEC in the majority of situations. However, extra care is needed in groups of sheep for which high levels of FEC aggregation might be expected. The implications of statistical error in FEC estimates depend on how they are used. The simulation-based approach presented here is a powerful tool for investigating the risks of error in FEC-driven treatment decisions in different situations, as well as for the statistical analysis of parasitological data in general.