The Health Belief Model (HBM; Rosenstock, 1966) was constructed to explain which beliefs should be targeted in communication campaigns to cause positive health behaviors. The model specifies that if individuals perceive a negative health outcome to be severe, perceive themselves to be susceptible to it, perceive the benefits to behaviors that reduce the likelihood of that outcome to be high, and perceive the barriers to adopting those behaviors to be low, then the behavior is likely for those individuals. A meta-analysis of 18 studies (2,702 subjects) was conducted to determine whether measures of these beliefs could longitudinally predict behavior. Benefits and barriers were consistently the strongest predictors. The length of time between measurement of the HBM beliefs and behavior, prevention versus treatment behaviors, and drug-taking regimens versus other behaviors were identified as moderators of the HBM variables' predictive power. Based on the weakness of two of the predictors, the continued use of the direct effects version of the HBM is not recommended.