The time and expense of sample processing is a bottleneck for many biomonitoring projects. Computer vision methods for pattern recognition have made enormous advances in recent years, and could become a powerful tool for automated insect identification. However, benthic macroinvertebrates pose special challenges: most taxa are nonrigid and thus variable in shape; some taxa are difficult to separate using superficial characters; and size and coloration may change during ontogeny. We developed an integrated method that includes machinery for automatically handling and imaging specimens under standardized conditions, and computer identification using concatenated histograms of local appearance features. Experiments using the stoneflies Calineuria, Doroneuria, Hesperoperla, and Yoraperla gave four-class accuracy of 82% and three-class (Calineuria and Doroneuria pooled) accuracy of 95%. Our method distinguished between Calineuria and Doroneuria as accurately as entomologists who were shown the same images (79.0% vs. 78.6%, respectively). Ongoing experiments include distinguishing among 8 stonefly taxa, and 10 non-stonefly “distractor” taxa that cover a range of morphological diversity from baetid mayfly to glossosomatid caddisfly. Our results suggest that automated methods could be viable for some sorting and taxonomy tasks, such as rapid identification and enumeration of target taxa.