We have studied the mechanism of contour perception by recording from neurons in the visual cortex of alert rhesus monkeys. We used stimuli in which human observers perceive anomalous contours: A moving pair of notches in 2 bright rectangles mimicked an overlaying dark bar. For control, the notches were closed by thin lines so that the anomalous contours disappeared or half of the figure was blanked, with a similar effect. Orientation-selective neurons were studied. With the receptive fields centered in the gap, 23 of 72 (32%) neurons tested in area V2 responded to the moving "bar" even though the stimulus spared their response fields, and when the notches were closed, their responses were reduced or abolished. Likewise, when half of the figure was removed, the neurons usually failed to respond. Neurons with receptive fields within 4 degrees of the fovea signaled anomalous contours bridging gaps of 1 degree-3.5 degrees. The anomalous-contour responses were compared quantitatively with response field profiles and length-summation curves and found to exceed the predictions by linear-summation and summation-to-threshold models. Summation models also fail to explain the effect of closing lines which add only negligible amounts of light. In V1, only one of 26 neurons tested showed comparable responses, and only with a narrow gap. The others responded only when the stimulus invaded the response field and did not show the effect of closing lines, or failed to respond at all. The contour responses in V2, the nonadditivity, and the effect of closure can be explained by the previously proposed model (Peterhans et al., 1986), assuming that the corners excite end-stopped fields orthogonal to the contour whose signals are pooled in the contour neurons.