We produced and tested rules to predict undervaccination among preschool-age emergency department (ED) patients. Data were gathered on demographics, vaccination status, health status, and health care utilization from parents, ED physicians, and ED charts at an urban teaching hospital in Rochester, N.Y. Primary care charts were reviewed to verify vaccination status. Using recursive partitioning, we developed decision rules to predict undervaccination. Decision rules were developed on a sample of 602 ED patients 4 to 48 months of age and then prospectively tested on 1832 ED patients aged 6 to 36 months. Factors associated with undervaccination for any vaccine included parental report of vaccination delay (odds ratio = 8.1; p < 0.001), inability to report the receipt of the appropriate number of vaccines (odds ratio = 4.5; p < 0.001), lack of health insurance (odds ratio = 3.6, p < 0.001), elapsed time since the last visit to primary care provider (p < 0.001), household size (p < 0.001), and maternal age (p < 0.01). Eight decision rules were produced that varied in their number of questions (one to six), sensitivity (0.27 to 0.87), and specificity (0.54 to 0.98). No single rule was both highly sensitive and highly specific. The rules' sensitivities and specificities were similar for the validation sample of 1832 patients. Thus a decision rule could not be produced that was both sensitive and specific. Identification of undervaccinated children by means of information available at an ED visit is inherently difficult. Interventions in the ED may be inefficient unless better methods of assessing vaccination status can be developed.