Many species have demonstrated the capacity to learn abstract concepts. Recent studies have shown that the quantity of stimuli used during training plays a critical role in how subjects learn abstract concepts. As the number of stimuli available in the training set increases, so too does performance on novel combinations. The role of set size has been explored with learning the concept of matching and same/different but not with learning the concept of difference. In the present study, pigeons were trained in a non-matching-to-sample task with an initial training set of three stimuli followed by transfer tests to novel stimuli. The training set was progressively doubled eight times with learning and transfer following each expansion. Transfer performance increased from chance level (50 %) at the smallest set size to a level equivalent to asymptotic training performance at the two largest training set sizes (384, 768). This progressive novel-stimulus transfer function of a non-matching (difference) rule is discussed in comparison with results from a similar experiment where pigeons were trained on a matching rule.