The perception of likeness is practically very much bound up with that of difference.That is to say, the only differences we note as differences, and estimate quantitatively,and arrange along a scale, are those comparativelylimited differences which we find between members of a common genus. (James, 1890, Vol. 1, p. 528) How are differences generated and why are certain differences noticed and others are not? The answer may depend on whether or not differences bear any relationship to commonalities. According to mental distance models of similarity (Nosofsky, 1987;Shepard, 1974;Shoben, 1983), the degree of difference is the inverse of the degree of similarity on any given dimension. In independent feature models (e.g., Tversky, 1977), differences are independent of commonalities; the differences between a pair of objects are simply any elements of the objects' feature sets that do not match. Although commonalities and differences may be differentially weighted according to the task or context, there is no necessary relationship between the common features and the psychologically salient differences.In contrast to the above accounts, the structural alignment approach (Gentner, 1983;A. B. Markman & Gentner, 1993a, 1993bMedin, Goldstone, & Gentner, 1993) posits representations composed of interconnected structures, rather than independent features or dimensional spaces. According to this view, as discussed below, differences are noticed relative to commonalities-that is, one first notices commonalities and then the differences that are related to those commonalities (e.g., whales and fish are both swimming creatures, but one has lungs, the other has gills).In the present series of experiments, we demonstrate four findings that link the structural alignment process with the generation of differences. First, differences are easier to generate for word pairs that have been recently aligned than for those that have not, demonstrating that structural alignment facilitates noticing differences. Second, this facilitation is specifically related to structural alignment and not merely the result of joint activation of the word pairs. Third, the degree of difference facilitation reflects the quality and extent of the pairs' common system. Fourth, pairs with deep and rich alignments elicit not only more differences, but more differences specifically related to the commonalities than do pairs with more sparse alignments.Structural alignment theory is a generalization of the structure-mapping theory of analogical reasoning (Falkenhainer, Forbus, & Gentner, 1989;Gentner, 1983), according to which comparison is accomplished by a process of alignment of structured representations of the entities or scenes being compared and the subsequent projection of inferences (Gentner & Markman, 1994Goldstone, Medin, & Gentner, 1991 High-similarity concept pairs that elicit many commonalities also elicit many related differences (Gentner & Markman, 1994;A. B. Markman & Gentner, 1993a, 1993b, 1996A. B. Markman & Wisniewski, 1997). This findi...