Origin–destination (OD) visualizations can help to understand movement data. Unfortunately, they are often cluttered due to the quadratic growth of the data and complex depictions of the multiple dimensions in the data. Many domain experts have designed visualizations to reduce visual complexity and display multiple data variables. However, OD visualizations have not been well classified, which makes it hard to employ such methods for reducing the visual complexity systematically. In this article, we propose a novel classification scheme for static OD visualizations that considers five aspects: the granularity of flows, the dimensionality in and of the display space, the semantics of the display space, the representation of nodes and flows, and the ways of relating two visualizations. We evaluate the proposed classification scheme using published visualization examples and show that it is effective and expressive.