In Experiment 1, we used six procedures in a series of unsuccessful attempts to obtain relationallearning using trial-unique pictorial stimuli in pigeons. The Experiment began by testing conventional (three-key) matching-to-sample (MTS)and nonmatching-to-sample (NMTS);in subsequent stages of the experiment we progressively incorporated features of techniques that do obtain relational learning in a single-key apparatus. In Experiment 2, we found that acquisition of NMTS using pictorial stimuli proceeded no more rapidly than acquisition of a conditional discrimination, Experiment 3 showed that acquisition of NMTS was more rapid than acquisition of MTSwhen plain colored stimuli were used, but not when pictorial stimuli were used. These three experiments suggest that pigeons do not recognize pictorial stimuli shown on different keys. In Experiment 4, between-key recognition was obtained with familiar but not with novel pictorial stimuli. It is argued that perceptual learning facilitates the detection of the between-key identity of complex stimuli, and that perceptual processes may underlie the difficulty in demonstrating relationallearning in pigeons.Pigeons find it very difficult to respond appropriately according to whether two simultaneously presented stimuli are the same or different. In matching-to-sample (MTS) or nonmatching-to-sample (NMTS) tasks, pigeons are able to learn to peck the side-key stimulus that is the same as (or different from) the sample stimulus exposed on the center key. However, only two alternative stimuli are used in conventional (N)MTS tasks, and when pigeons are subsequently tested with two novel stimuli, little or no evidence of positive transfer is obtained (e.g., Wilson, Mackintosh, & Boakes, 1985a). Such findings suggest that pigeons, rather than achieving relational learning (choosing, that is, according to the relationship, same or different from sample, that holds for the choice stimuli), solve these tasks as conditional "if-then" discriminations (selecting one stimulus when a given sample is shown and the other when the alternative sample is shown). When trial-unique stimuli (stimuli that are shown at the most only once per session) are used, a solution in terms of (multiple) conditional discriminations becomes very difficult, and pigeons have been shown to achieve relational learning, showing significant This research was supported by a grant from the U.K. Medical Research Council. M.G. is now at the Department of Pharmacology, University of Edinburgh Medical School, I George Square, Edinburgh EH8 9JZ. R.C.H. is now at the Sub-Department of Animal Behaviour, Madingley, Cambridge CB3 8AA. Correspondence concerning this article should be addressed to E. M. Macphail, Department of Psychology, University ofYork, York, YOI 500, England (e-mail: emml@ york.ac.uk).-Accepted by previous editor, Vincent M. LoLordo 83 positive transfer to entirely novel stimuli in a samenessdifference discrimination as well as an MTS task (Santiago & Wright, 1984; Wright, Cook, Rivera, Sands, & Delius, 1...