like the duck/rabbit and the snail/elephant are reversible. We distinguish between two types of reversal: those that entail a change in reference-frame specification as well as a reconstrual of image components (reference-frame realignments) and those that entail reconstruals only (reconstruals). We show that reference-frame realignments can occur in imagery, particularly ifobservers are given an explicit or an implicit suggestion; and that reconstruals of images occur commonly, regardless of experimental conditions. In addition, we show that images constructed from good parts are more likely to reverse than images constructed from poor parts. On the basis of these results, we propose a functional organization of shape memory that is consistent with shape recognition findings as well as with our reversal findings.Given that imagery theories address questions regarding the nature of the imagined representation, theories of shape recognition are implicit in theories of imagery (see Kosslyn, Pinker, Smith, & Shwartz, 1979). Indeed, because imagined shapes must be generated by memory structures, studies of the attributes of mental images (henceforth also just "images") may yield informationabout thecharacteristics of memory representations of shape. This issue underlies the current debate about whether mental images can be ambiguous-that is, whether such images can be reversed, reconstrued, or reinterpreted,just as picturescanbe.Some philosophers (e.g., Casey, 1976;Fodor, 1981) and psychologists (e.g., Kolers, 1983) have argued that mental images cannot be ambiguous, but empirical data have been lacking until recently. Chambers and Reisberg (l985) found that not 1 of 35 observers who originally saw only one of the two potential interpretations of the duck/rabbit ambiguous figure (Jastrow, 1900) could find the alternative interpretation in their memory image of the figure. Hence, they concluded that mental images of ambiguous pictures refer unambiguouslyto only one potential interpretation of the picture. Further, they argued that in the process of creating images, observers access shape representations by means of semantic representations, and that the two cannot be decoupled: Having seen a rabbit in a picture, one subsequently accesses a mental image of a rabbit that cannot be transformed into an image of a duck. According to this view, those of us who are familiar with the duck/rabbit figure, and might suppose that we are able to imagine a reversal, are simply replacing one image with another, rather than discovering an alternative interpretation of a single mental image.Although Chambers and Reisberg's (1985) findings were quite striking, Finke, Pinker, and Farah {l989} demonstrated that semantic and structural representations of objects can be separated. When asked to manipulate and combine alphanumeric characters or simple geometric shapes in imagery, their observers discovered a number of new shapes in the resultant configurations: for example, a particular arrangement of the letters J and D was interpreted as...