Subjects judged whether two marks placed at different positions along a curved contour were physically the same. When targets were separated by a concave curvature extremum--corresponding to a part-boundary--decision latencies were longer than when they straddled an equally curved convex extremum, demonstrating a "single-part superiority effect". This difference increased with both stimulus duration and the magnitude of contour curvature. However, it disappeared when the global configuration was not consistent with a part-boundary interpretation, suggesting a critical role of global organization in part decomposition.