People can decide whether an image depicts a flat, painted surface or a shaded, 3-dimensional one. We sought a computational model to account for these judgements of shading and reflectance. We generated 60 test images. 18 subjects used a 5 point scale to rate how likely each image was to have been made by a painter, making marks on a flat surface, or a sculptor, creating a 3-dimensional object with no marks on it. In our computational model, we assumed that each image was either all shape or all paint. We used a linear shading model to render shapes. In our Bayesian approach, we assigned a prior probability to each possible shape and paint configuration, penalizing the absolute value of spatial derivatives of shape and paint. For each image, we found the most probable all-shape explanation. We formed a shapeness index based on the ratio of the probability of the best shape solution to the probability of the all-paint solution. Images judged to have been painted can be created by shape and lighting, but require large areas of sloped surfaces, which the prior probability term penalizes. These shape explanations also require precise (coincidental) alignment with the assumed light direction. We found a 0.7 correlation between our shapeness index and the image ratings of the subjects, showing that our computational model accounts reasonably well for subject judgements of shading and reflectance.