In their understandable desire to avoid the rigidity of some classification schemes, Romeijn and van Loo describe an empirically driven system for classification that emphasizes black-box prediction over questions of reduction or realism. I note that belief in diagnostic entities seems to persist even in a theoretical domain that is a-reductionist, and wonder why. The problem, I note, is very similar to the one faced by MacCorqodale and Meehl more than fifty years ago, when they were trying to extract clinical psychology from the tight strictures of operationalism.MacCorquodale and Meehl’s “hypothetical constructs” are a-realist in the same sense that Romeijn and van Loo’s prediction models are a-reductionist.