1 Introduction The process of contour integration has been intensely studied since the early 90s, with several different psychophysical tasks. Field and colleagues (1993) investigated the ability to detect a path made up of Gabor elements, embedded in a field of randomly oriented Gabors, as function of the spatial separation and the relative orientation of the elements of the path. Performance in the task was found to decrease as the spatial separation among the elements increased, and also as the orientation of adjacent elements of the path deviated away from collinearity. Findings were interpreted as evidence for the existence of an `association field' around each oriented contour element, which can be conceived as a set of local orientation signals radiating outward in every direction from an oriented edge (Shipley and Kellman 2003). The theory of the association field maintains that contour integration is mediated by lateral connections between orientation-sensitive neurons in the early visual areas, and it has been related to the results of several studies of contrast detection (Wehrhahn and Dresp 1998; Dresp 1999; Tzvetanov and Dresp 2002) and orientation discrimination (Brincat and Westheimer 2000), which showed that these tasks can be facilitated (and the relative thresholds lowered) by the presence of collinear elements (flanks) next to the target location. Both kind of studies suggested the existence of two distinct spatial regimes of interaction: a short-range domain, in which facilitation is sensitive to stim- ulus features such as contrast polarity and intensity, and a long-range domain, in which facilitatory effects are found regardless of changes in those features (Field et al 2000). The outer limit of the long-range domain for collinear elements has been estimated around 2.5 deg of visual arc, while the one of the short-range domain was found to be between 15 min of arc (in orientation-discrimination studies öBrincat and Westheimer 2000) and 25 min of arc (in line ^ contrast detection studies öTzvetanov and Dresp 2002) in the direction coaxial with the elements, and around 3 ^ 5 min of arc in the orthogonal direction. Features of the selectivity for contrast polarity in contour integration revealed by a novel tilt illusion Perception, 2011, volume 40, pages 1357 ^ 1375 Stefano Guidi, Oronzo Parlangeli, Sandro Bettella½, Sergio Roncato½ Communication Sciences Department, University of Siena, p.zzo S. Niccolo© , via Roma 56, 53100 Siena, Italy; e-mail: stefano.g73@gmail.com; ½ Department of Psychology, University of Padua, via Venezia 10, 35131 Padua, Italy Received 20 December 2010, in revised form 12 October 2011; published online 14 December 2011 Abstract. We studied a novel illusion of tilt inside checkerboards due to the role of contrast polarity in contour integration. The preference for binding of oriented contours having same contrast polarity, over binding of opposite polarity ones (CP rule), has been used to explain several visual illusions. In three experiments we investigated how the binding effect is influenced by luminance contrast value, relatability of contour elements, and distance among them. Experi- ment 1 showed that the effect was indeed present only when the CP rule was satisfied, and found it to be stronger when the luminance contrast values of the elements are more similar. In experi- ment 2 the illusion was reported only with relatable edges, and its strength was modulated by the degree of relatability. The CP-rule effectiveness, thus, seems to depend on good continuation. The intensity of contrast polarity signals propagating from an oriented contour might be the less intense, the more its direction deviates from linearity. In experiment 3 we estimated the distance threshold and found it to be smaller than the one found for other illusions, arising with collinear fragments. This seems to show that the reach of the contrast polarity signal inside the association field of a contour unit is shorter along non-collinear orientations than along collinear ones. doi:10.1068/p6897