I propose that strong claims about the superiority of latent variable (LV) structural modeling, compared to other causal approaches to nonexperimental data, are both overstated and premature. Focusing primarily on a recent article by Huba, Wingard, and Bentler (1981) appearing in this journal, I argue both that the authors' application of these models is seriously flawed and that the data employed to test the models are of questionable quality. I conclude that statements about powerful inferences that can be made from analyses of LV models should be moderated. A more prudent position is advocated in this paper: I argue that the evaluation and testing of LV models should proceed, but with caution and self-criticism; even highly sophisticated methods like LV modeling cannot substitute for the collection of high quality data and the clear operationalization of constructs and hypotheses.