Nonclinical testing to predict the abuse potential of central nervous system (CNS) drug candidates is a mandatory part of the safety pharmacology assessment for medications seeking approval for human use. In the "standard model," the drug candidate is tested to determine whether its psychoactive effects generalize to the discriminative cue of an abused drug that animals have been trained to recognize. However, CNS drugs with novel pharmacological mechanisms are challenging, and in response, the regulatory agencies have recommended alternative experimental designs. Variant 1: test the drug candidate in a series of drug-discrimination experiments that exemplify the major classes of abused drugs. Variant 2: use the drug candidate as a training cue. Back-test examples from established classes of abused drugs to see if they generalize to the drug candidate's cue. We critically assessed the pharmacological and translational validity of these protocols. The standard model is underpinned by decades of research and refinement and has the highest degree of translational validity. Question marks exist over the validity of substitution results when the drug candidate has no affinity for known abuse-related targets. Published research does not support the use of either of the alternative models. On the contrary, these models have no pharmacological rationale and, consequently, no translational validity. The review contains a decision tree on the appropriate application of the standard drug-discrimination model, together with recommendations for adapting the test when characterizing the psychoactive properties of drug candidates acting on novel CNS targets.