Visual memory holds in mind details of objects, textures, faces, and scenes. After initial exposure to an image, however, visual memories rapidly degrade because they are transferred from iconic memory, a high-capacity sensory buffer, to working memory, a low-capacity maintenance system. How does visual memory maintenance work? This dissertation builds the argument that the maintenance of short-term visual memories is analogous to the act of breathing: it is a dynamic process with a default behavior that explains much of its usual workings, but which can be observed, overridden, and controlled. Chapter 1 shows how the act of trying to remember more information causes people to forget faster and to remember less (“load-dependent forgetting” and “overreaching”). It then shows how the paradigm of evolution can be applied to the problem of maintenance, with memories competing over a limited memory-supporting commodity, explaining these effects. Chapter 2 presents experiments on metamemory, the ability of people to observe and make decisions about their own memories. The experiments isolate a component of metamemory that monitors a memory’s quality as it degrades over time. Chapter 3 connects memory to metamemory, drawing on work from reinforcement learning and decision theory to liken the problem of memory maintenance to that of an agent who sequentially decides what to prioritize in a partially observable mind.