Covariance structure modeling, also known as structural equation modeling or causal modeling, appears increasingly popular. Such techniques can be used to conduct tests of complex theory on empirical data. To conduct such tests, researchers need measures of known factor structure and the knowledge of structural relations among the constructs of interest. Researchers typically have neither the required measures nor the knowledge of structural relations. Instead of conducting tests of theory, researchers use covariance structure models to develop measurements and theoretical models. This paper discusses why such use of covariance structure models is unlikely to produce scientific progress and proposes some alternative procedures thought to be more fruitful.