Working memory is essential for organisms to solve problems related to their survival and to adapt to changes in their environment. Researchers sought to create a non-human model of working memory that could be used to better understand its predictive value and underlying brain function. Several of these studies were conducted using the odor span task (OST) with rodents, and here, we present the first OST with domestic dogs (n = 6). The OST is an incrementing non-match-to-sample task in which dogs were presented with both a session novel (S +) and a previously encountered (S -) odor on each trial. A response to the session novel odor was always reinforced. Upon meeting training criterion on sessions with 24 trials or odors to remember, the dogs were tested on the OST with up to 72 odors to remember in the session. All dogs learned the OST and displayed accurate performance (≥ 79%) for the largest set size of 72 odors. In an analysis focused on the effect of intervening odors (i.e., the number of trials since the S - was last encountered), dogs demonstrated above-chance performance for up to eight intervening odors. The implications of these findings are discussed in the context of dog working memory for odors.