Most tests of cognitively oriented theories of health behavior are based on correlational data. Unfortunately, such tests are often biased, overestimating the accuracy of the theories they seek to evaluate. These biases are especially strong when studies examine health behaviors that need to be performed repeatedly, such as medication adherence, diet, exercise, and condom use. Several misleading data analysis procedures further exaggerate the theories' predictive accuracy. Because correlational designs are not adequate for deciding whether a particular construct affects behavior or for testing one theory against another, most of the literature aiming to test these theories tells us little about their validity or completeness. Neither does the existing empirical literature support decisions to use these theories to design interventions. In addition to discussing problems with correlational data, this article offers ideas for alternative testing strategies.