Abstract
Cronbach's alpha coefficient is one of the most widely used indices of the quality of a psychometric test, a popularity that has also earned it an abundance of criticism and sustained competition from a large number of candidates. Do these coefficients assess the "reliability" of the test or its "internal consistency", what do these designations mean, and how do the interpretations given to these indices relate to their mathematical structure? The historical and semantic drift of the term "reliability" and the obvious vagueness of the term "internal consistency" are the probable causes of this confusion. We examine these different questions in the light of Cronbach's 1951 article and try to set the record straight by advising the researcher to define a priori the quality of the test he is trying to verify and to ensure that the calculation procedure he chooses corresponds explicitly to this quality.