A key skill in early human development is the ability to comprehend communicative intentions as expressed in both nonlinguistic gestures and language. In the current studies, we confronted domestic dogs (some of whom knew many human ‘words’) with a task in which they had to infer the intended referent of a human's communicative act via iconic signs – specifically, replicas and photographs. Both trained and untrained dogs successfully used iconic replicas to fetch the desired item, with many doing so from the first trial. Dogs’ ability to use photographs in this same situation was less consistent. Because simple matching to sample in experimental contexts typically takes hundreds of trials (and because similarity between iconic sign and target item did not predict success), we propose that dogs’ skillful performance in the current task reflects important aspects of the comprehension of human communicative intentions.