The identification of functional feeding traits in aquatic macroinvertebrates often requires a morphology-based identification of species, which is important for trait-based methods of biological assessment. The extent of functional homogenization is compared along scales of impairment, where trait-based information is used as an input in models that examine degradation pathways. However, trait-based information is not always readily available for all groups of aquatic insects, especially for species diverse families, such as chironomids (Diptera: Chironomidae). Taxonomic challenges and ambiguous traits complicate the use of chironomid larvae in trait-based bioassessment. Here, we examine the use of geometric morphometric analysis (GMA), deep learning (Convolutional Neural Networks), and computer vision (deep CNN) applied to the mouthparts (mandibles) of chironomid larvae as a proxy for identifying the relationship between the functional morphology and food acquisition behaviour. We determined the variability in morphology of mandibles for 23 taxa of chironomid larvae from different genera, subfamilies, and their Functional Feeding Group (FFG). Analysis using GMA showed that the five different FFGs examined had different mandibular traits that significantly varied in shape and size. A deep CNN model was then built that was able to classify the 23 taxa into their respective FFG automatically with 92.31 % accuracy. A gradient-weighted Class Activation Mapping (Grad-CAM) algorithm found that the most important part of mandibles for classification were the gula and mandibular joint. We introduced three additional species to the deep CNN models to test whether automatic classification would directly and automatically identify traits of the specimens independently from taxonomic identification. The deep CNN process avoids issues surrounding both taxonomic identification and previous knowledge of a specific taxon’s feeding trait, and in all cases the model classified taxa correctly based on their mandibular traits. The use of deep learning approaches could substantially enhance the use of trait-based approaches and increase the reliability and use of chironomids in bioassessment.