This is in line with comparable differences in context generalization that have been observed in classical conditioning experiments (see Bouton, 2004). At first sight, Experiment 2 provided support for this asymmetry: the gradient of acquisition generalization was flat on the first test trial, whereas a sharp and orderly gradient of extinction generalization was observed at final test. Hence, even with complete initial generalization of acquisition, generalization of extinction back to the S+ was weak (evidenced by recovery of outcome-expectancy to the S+). However, as we have seen in the section above, the flat gradient on the first test trial may have been confounded by uncertainty in this study.