One of the fundamental issues in lightness perception concerns the structure of lightness representations. A central debate involves the role of layered image representations in the computation of surface lightness. Barrow & Tenenbaum (1978) suggested that the visual system performs an analysis in which the image is decomposed into a series of layered maps that represent the illumination, shape and lightness of surfaces in the image. More recently, a number of authors have questioned the role of layered image representations in the computation of surface lightness (Gilchrist et al., 1999; Adelson, 2000). We present a new class of lightness illusions that definitively demonstrate the role of layered image representations in the computation of surface lightness. We generated random patches of texture with a power spectrum that varies as 1/f^^n (n>2). The spatial frequency components were then summed with random phases and orientations. Three versions of each image were constructed with this seed image: a higher contrast (HC) image; an image in which the maximal luminance in the texture fell below the mean of the HC display (a dark texture); and an image in which the minimal luminance was above the mean luminance of the HC image (a light texture). A small circular central patch of the HC image was placed on the corresponding position on both the dark and light textures. Both HC targets appeared as homogeneous circular discs overlaid with a transparent layer. Observers matched the perceived lightness of the central HC patch on each background. Our results revealed the largest lightness illusion reported to date: the HC target on the light background appeared black, and the HC target on the dark background appeared white. We show that the contrast relationships between the target patch and the surround are critical in causing these effects, particularly the polarity relationships. These results demonstrate that layered image representations can play a critical role in determining perceived surface lightness.