This article analyzes Dillon and Kumar’s reanalysis of data previously reported by Fishbein and Ajzen (1974) and subsequently reanalyzed by Bagozzi and Burnkrant (1979). It also performs new analyses of the same data addressing the construct validity of the unidimensional and the two component attitudinal models. We show that the unidimensional attitude model fails to achieve convergent validity and the two component model achieves convergent, discriminant, and predictive validity. Further, we point out that Dillon and Kumar misinterpreted and did not pursue their analyses far enough, and they failed to present anomalous results tending to disconfirm their hypotheses.