An attempt is made to indicate why the concept of "true score" naturally leads to the belief that test validity must increase with an increase in test and/or average item reliability, and why this is correct for the classical single-factor model first introduced by Spearman. The statistical model used by Loevinger is introduced to establish the "attenuation paradox", and, in intuitive terms, attempt to explain why the "attenuation paradox" holds in this particular model. This is accomplished by showing that high (internal) consistency or reliability of test scores is an asset in increasing test validity under the classical single-factor statistical model for mental tests, but can be a liability when item scores are modelled as in the statistical model discussed by Loevinger. It is hoped that by this exposition, mental test specialists will be led to more critical appraisal of commonly used techniques and concepts (including the "corrections for attenuation"), and will check that their methods of test construction and comparison are consistent with their statistical models. (Author/PR)