Cognitive ethology cannot be done well unless its proximate philosophical underpinnings are got straight; this paper tries to help with that. Cognitive attributions are essentially explanatory—if they did not explain behavior, there would be no justification for them—but it doesn’t follow that they explain by providing causes for events that don’t have physical causes. To understand how mentalistic attributions do work, we need to focus on the quartet: sensory input, belief, desire, and behavioral output. We also need to be able to study classes of sensory inputs—one-shot deals are uninterpretable. The crucial guiding rule is, roughly: The animal’s behavior shouldn’t be explained by attributing to it the belief that P unless the behavior occurs in sensory circumstances belonging to a class whose members are marked off in some way that involves the concept of P and not in any way that is lower than that. The higher/lower distinction can be understood so that the guiding rule is helpful not only in deciding what thoughts to attribute to an animal but also in deciding whether to attribute any thoughts at all.