Humans can discriminate one visual contour from another on the basis of small differences in orientation. This capability depends on cortical detectors that are selective for a small range of orientations. We have measured this orientation bandwidth and the suppression that helps to shape it, with a reverse correlation technique. Human subjects were presented with a stream of randomly oriented gratings at a rate of 30 per second. Their task was to press a key whenever they saw an orientation nominated as the target. We analyzed the data by finding the probability density of two orientations: One preceded the key-press by the reaction time, and the second preceded the first by up to 100 ms. The results were as follows: (1) One grating facilitated the following one in producing a key-press when the gratings differed little in orientation. The estimate of orientation bandwidth resulting from this facilitation was 38°. (2) A large angle between the two orientations reduced the probability of a key-press. This finding was best modelled as a suppression that did not vary with orientation, consistent with the idea that cross-orientation suppression is non-oriented. (3) Analysis of non-consecutive grating pairs showed that cross-orientation interactions lasted no longer than 67 ms.