Rescorla (2000) devised the compound test procedure as a means of comparing changes in associative strength when cues with different training histories are conditioned in compound. It was specifically intended to dissociate changes in learning from changes in performance, and thereby, permit inferences about learning independently of assumptions regarding how learning translates into performance. In an elegant series of studies, Rescorla (2000, 2001) used this procedure to show that cues conditioned in compound undergo unequal associative change such that the poorer predictor of the outcome undergoes greater change rather than the equal change predicted by theories (e.g., Rescorla & Wagner, 1972) that rely on a common error term. Rescorla explained the data from the compound test procedure by proposing that associative change is calculated using a combination of two error terms, a common error term that carries the predictions of all cues present on a trial and an individual term that carries the prediction of any cue in isolation. This article is in two parts. The first used simulations to show that a theory, such as Rescorla-Wagner, which just relies on a common error term, can explain the compound test data if the function that translates learning into performance is double-sigmoidal across the full range of associative strength (i.e., from inhibition through to excitation). The second part likewise used simulations to show that a theory, such as the comparator theory (Miller & Matzel, 1988), which does not invoke a common error term, can also explain the compound test data. Thus, a common error term is sufficient to explain the compound test data, but it is not necessary. (PsycINFO Database Record (c) 2019 APA, all rights reserved).