We recorded responses of single neurons in V1 and V2 of anesthetized macaque monkeys to brief (100 ms), rapid (5/s) random sequences of these images. The stimuli were matched in mean luminance and were presented against a gray background in a window of 4 deg diameter, which was larger than the classical receptive field of any of the neurons. In both areas, neurons responded more to some texture images than others. In V1, as expected, naturalistic and noise images evoked similar responses. In V2, however, almost all neurons responded more vigorously to the naturalistic images than to the corresponding noise images, and this difference was evident throughout the response. A late part of the V1 population response (about 100 ms after response onset) showed a weak but significant enhancement for naturalistic images. We also examined how well textures could be identified from the population responses in each area. Classification performance in V1 was similar for naturalistic and noise images, but in V2 performance was better for naturalistic images than for noise images, perhaps because of the difference in response rates. We also found that generalization of classification across samples within a texture category was significantly better in V2 than in V1. These results suggest that selectivity for the image properties captured by the texture model, as well as invariance to the properties discarded by the model, increase as signals pass from V1 to V2.