Saccade adaptation is a form of motor learning that maintains saccade accuracy when facing new sensorimotor contingencies. Researchers have shown that distinct saccade gain states can be adapted simultaneously in a double-step paradigm depending on contexts acting on the motor command (e.g. velocity or direction of target motion, head tilt, orbital eccentricity, or vergence). However, purely visual cues such as target color or shape consistently failed to drive different gain states during saccadic adaptation. Since Deubel's work in 1995, this question has remained unsolved despite extensive research in both human and non-human participants. The present work investigates whether this lack of contextual control might be due to a lack of relevance of the contextual cue to the saccadic system. We modified the conventional intrasaccadic step paradigm by displaying a distractor at the back-stepped location when the target stepped forward and vice-versa. In a first experiment the pre-saccadic target color (red or green) indicated its intra-saccadic step (backward or forward) and a distractor (green or red) was displayed after the saccade landed (at the forward-stepped or back-stepped location). We observed contextual control of saccade adaptation, which did not occur when the distractor was not displayed. In two other experiments we found the same effect using shape or shape and color as contextual features. We conclude that feature-based selection of the target promotes contextual adaptation. Three additional experiments further confirmed that i) when two features of the context cues are conflicting, contextual adaptation is blocked, ii) the spatial reference provided by the distractor is not responsible for our contextual adaptation and iii) removing the distractor during the adaptation phase induces a progressive decrease in adaptation. These results indicate that any context could potentially control contextual saccadic adaptation provided that the relation between the context cue and the saccadic response is made relevant. Meeting abstract presented at VSS 2016