Fox squirrels have an impressive ability to remember the location of stored food. In doing so, they combine information from landmarks of different types. We deﬁne a Bayesian model that indicates how an ideal observer would optimally integrate landmark cues, and fit this model to the decisions made by squirrels in a spatial memory task. The resulting model provides a unifying framework for characterizing different strategies to cue integration, and a tool for investigating the circumstances under which particular cues are used. We show that the best fitting models changed depending on the season at testing and the details of the task. These analyses support previous claims that squirrels adopt flexible strategies in landmark use.