Network analysis provides a unified framework for investigating different types of species interactions at the community level. Network analysis is typically based on null models that test for specific patterns in network topology. Here we use a novel predictive approach to investigate the topology of a mistletoe–host network. It has been hypothesised that Australian mistletoes mimic the phenotype of their preferred hosts to avoid herbivory. We developed a deterministic model based on phenotypic similarity to predict the topology of a quantitative network between Lauranthaceaous mistletoes and their hosts. We quantified mistletoe–host interactions in a semi-arid woodland central Australia, along with the size, shape and colour of leaves produced by both players in the interaction. Traditional null model analyses showed support for negative co-occurrence patterns, web specialisation and strong links between species pairs. However, our deterministic model showed that the observed network topology could not be predicted by phenotypic similarity, suggesting that Australian mistletoes do not mimic their hosts.