We examined the phylogenetic relationships between species and genera within the caddisfly subfamily Drusinae (Trichoptera: Limnephilidae) using sequence data from two mitochondrial loci (cytochrome oxidase 1, large subunit rRNA) and one nuclear gene (wingless). Sequence data were analysed for 28 species from five genera from the subfamily. We analysed individual and combined data sets using a Bayesian Markov Chain Monte Carlo and a maximum parsimony approach and compared the performance of each partition for resolving phylogenetic relationships at this level. In terms of resolution and phylogenetic utility wingless outperformed the two mitochondrial gene partitions. Using both Shimodaira-Hasegawa and expected likelihood weights tests we tested several hypotheses of relationships previously inferred based on adult morphological characters. The data did not support the generic concept, or many previously proposed species groupings, based on adult morphology. In contrast, the molecular data correlated with the morphology and feeding ecology of larvae. Using Bayesian ancestral character state reconstructions we inferred the evolution of feeding ecology and relevant larval morphological characters. Our analyses showed that within the subfamily Drusinae two derived feeding types evolved. One of these--grazing epilithic algae--is otherwise unusual in the Limnephilidae and may have promoted the high degree of diversity in the Drusinae.