The relational account could first be considered in random conditions. Actually, an analysis of RTs according to the preceding trial, evidenced some perceptual priming effects in the random conditions. The RTs were faster when the target was a similar item in the previous trial, than when it was a more different item7. A relational target template might have led to the results observed in the random conditions. Several arguments, however, went against this interpretation. First, as argued, no attentional set was expected in these conditions, since no information about the target was available, nor useful. A relational template would not have been more relevant than a feature-based template. Second, no inter-trial priming could account for the size effects observed in the random conditions, since the three squares (i.e. the small medium and large ones) were the target at the very same frequency in the random conditions. Finally, if this relational hypothesis accounted for the size effects in the random condition, then size effects should have been similar in the large- and small- relevant conditions, in Experiment 2. Yet the general size effects were much stronger in the large- than in the small-relevant condition. This difference suggested that an attentional set could not account for the whole story; Rather, salience effects were more likely responsible for the size effects observed in the random conditions. All in all, the relational hypothesis seems not able to account for these size effects. On the other hand, the observed similarity effects in priming were not contradictory to feature-based priming mechanism, as argued.
To assess the existence of inter-trial priming effects, Ulrich Ansorge suggested to us the following analysis. Three conditions of transition between trial n-1 to trial n were compared. In the "identical" condition, the same target was repeated (i.e. a large target then a large target). In the "proximate" condition, targets in trial n-1 and trial n were different but similar (i.e. a medium then a large target or a medium then a small target). Finally, in the "far" condition, targets differed more sharply (i.e., a small then a large or a large then a small). These effects were computed for small and large targets preceded by a similar relevance condition (i.e. random). The effects of similarity in priming appeared significant between "far" (1112 ms) and "proximate" (1051 ms) conditions in Experiment 1.1 (t(t(17) = 2.12, p < .025) and in Experiment 3 (block-wise; "far" :894 ms, "proximate": 846 ms, t(17) = 2.09, p < .026). On the contrary, in the Experiment 2 (trial-wise), this effect was not significant ("far": 1017 ms, "proximate": 999 ms, (t(17) = 0.70, p > .24).