Accurate exposure assessment remains a challenge in occupational epidemiology. We evaluated one approach, use of a job-exposure matrix (JEM), by applying the National Institute for Occupational Safety and Health (NIOSH) JEM to a large case-control birth defects study that included parental occupation information. We investigated the JEM exposure predictions in several ways and found that for a substantial proportion of the parents in the birth defects study, the JEM yielded either no exposure data or nonsense predictions. Among exposure predictions that were plausible, most were of low probability. The high probability exposure predictions were statistically unstable, and neither low nor high probability exposure predictions were reliable. There was considerable discrepancy between the JEM predictions and expert assessments for five exposures of interest. Application of the NIOSH JEM to the birth defects study database (and probably other databases as well) does not provide a useful means of assessing occupational exposures.