Explaining the exceptional diversity of herbivorous insects is an old problem in evolutionary ecology. Here we focus on the two prominent hypothesised drivers of their diversification, radiations after major host switch or variability in host use due to continuous probing of new hosts. Unfortunately, current methods cannot distinguish between these hypotheses, causing controversy in the literature. Here we present an approach combining network and phylogenetic analyses, which directly quantifies support for these opposing hypotheses. After demonstrating that each hypothesis produces divergent network structures, we then investigate the contribution of each to diversification in two butterfly families: Pieridae and Nymphalidae. Overall, we find that variability in host use is essential for butterfly diversification, while radiations following colonisation of a new host are rare but can produce high diversity. Beyond providing an important reconciliation of alternative hypotheses for butterfly diversification, our approach has potential to test many other hypotheses in evolutionary biology.