In two experiments ( ns = 3 plus a previously tested child, and 2, respectively), children learned delayed matching with complex samples, each consisting of a form and a printed nonsense word. Forms or printed words were comparison stimuli. For form comparisons, selecting the form identical to that in the preceding sample was reinforced. For printed word comparisons, selecting the word identical to that in the preceding sample was reinforced. During testing, the children then matched the forms and printed words to one another. In subsequent training, the samples were (a) old forms combined with new words or (b) old words combined with new forms. Novel matching performances among forms and words appeared across these training phases. Word-form contiguity in a matching-to-sample context may contribute to the formation of classes of stimuli that may be equivalent.