Two word-trained dogs were presented with acts of reference in which a human pointed, named objects, or simultaneously did both. The question was whether these dogs would assume co-reference of pointing and naming and thus pick the pointed-to object. Results show that the dogs did indeed assume co-reference of pointing and naming in order to determine the reference of a spoken word, but they did so only when pointing was not in conflict with their previous word knowledge. When pointing and a spoken word conflicted, the dogs preferentially fetched the object by name. This is not surprising since they are trained to fetch objects by name. However, interestingly, in these conflict conditions, the dogs fetched the named objects only after they had initially approached the pointed-to object. We suggest that this shows that the word-trained dogs interpret pointing as a spatial directive, which they integrate into the fetching game, presumably assuming that pointing is relevant to finding the requested object.