Identification of aquatic macroinvertebrates is a skill that requires considerable time and expertise. Being able to recognize key morphological features of these macroinvertebrates is accomplished through study, the ability to use dichotomous keys, and hours of practical experience with a microscope. Identifying aquatic macroinvertebrates is essential to catalog species, conduct population and community studies, and monitor aquatic systems. Because identifying aquatic macroinvertebrates is a specialized skill, many researchers send macroinvertebrate samples to taxonomic experts for identification. This practice costs time and monetary resources, both of which constrain the scope of biomonitoring studies. Our purpose for this project was to determine the accuracy of combined optical imaging and deep learning techniques, such as convolutional neural networks (CNNs), in macroinvertebrate taxonomic identification, and to address the technological gaps that have stymied the adoption of optic-related AI and deep learning in biomonitoring. Here, we hypothesize that the obstacles to this advancement are hardware-related (i.e. optical properties), rather than computational limitations of deep learning. In other words, we ask “If we have the best available, high-quality imagery at our disposal, how accurate can AI be in taxonomic identification?” To test this, we conducted a numerical experiment comparing the accuracy of CNNs to identify taxa across different sources of images, including highly controlled vs uncontrolled settings and with and without background behind the organisms. We manually collected high-quality imagery of 23 different families of Ephemeroptera, Plecoptera, and Trichoptera taxa with a stereomicroscope and a controlled illumination chamber to build a training library. The background of these images were removed using the “rembg” function in Python. We compared the accuracy performance of CNNs on our imagery database to another benchmark database of images collected under less controlled settings. We seek to understand the key aspects limiting the use of these technologies for aquatic macroinvertebrate identification.