Errors made by animals in memory paradigms are not always due to failure of memory.

It is commonly assumed that errors in animal memory paradigms such as delayed matching to sample, radial mazes, and food-cache recovery are due to failures in memory for information necessary to perform the task successfully. A body of research, reviewed here, suggests that this is not always the case: animals sometimes make errors despite apparently being able to remember the appropriate information. In this paper a case study of this phenomenon is described, along with a demonstration of a simple procedural modification that successfully reduced these non-memory errors, thereby producing a better measure of memory.