Medical waste, i.e. waste produced during medical activities in hospitals, clinics and laboratories, represents hazardous waste whose management involves special care and high costs. However, this kind of waste contains a significant fraction of highly valued materials that can enter a circular economy process. To this end, in this paper, we propose a computer vision approach for assisting in the primary sorting of medical waste. The feasibility of our approach is demonstrated on a representative dataset we collected and made available to the community, with which we have trained a model that achieves 100% accuracy, and a new dataset on which the trained model exhibits good generalization.