Texture synthesis models based on joint statistics of wavelet coefficients across scale, orientation, and position have become a popular tool for studying the representations that support texture processing in the human visual system. In particular, the summary statistics implemented in the Portilla-Simoncelli model account for observer performance in crowding, search tasks, and the response properties of V2 neurons. Presently, we chose to investigate whether or not this set of summary statistics was also sufficient to support performance in a texture discrimination task that required invariance to illumination. We selected 14 pairs of visually-matched textures from the Amsterdam Library of Textures, including a diverse range of material properties. Each texture was depicted under two illumination conditions: diffuse overhead lighting and strong side lighting. We generated a synthetic image from each original texture image using the Portilla-Simoncelli model. Using these images, we asked observers (N=13) to complete four match-to-sample tasks. Each task briefly presented (250ms) a sample texture to be matched to one of two test images presented after a 500ms ISI- one test image depicted the same texture as the sample, the other a visually-matched distractor. In the Illumination-change condition, the target texture and the sample were differently illuminated; In the No-change condition, lighting was the same. Observers completed both conditions using real and synthetic textures in separate blocks. We observed significant main effects of real/synthetic appearance (p<0.001) and illumination condition on accuracy (p<0.001), qualified by an interaction between these factors (p<0.001), such that performance in the No-change condition was slightly worse for synthetic textures (~5% cost), but that this difference was much larger in the Illumination-change condition (~20% cost). We conclude that invariant texture recognition relies on statistics not included in the P-S model and natural texture appearance overall leads to better discrimination performance. Meeting abstract presented at VSS 2014