BACKGROUND: Although the risk of vaccine-associated paralytic poliomyelitis (VAPP) has remained relatively constant during the past 30 years, estimates of VAPP depend largely on the completeness of reporting to the existing passive surveillance system. The National Vaccine Injury Compensation Program constitutes an alternative system for reporting VAPP, and data available from this system permitted us to evaluate the completeness of the national poliomyelitis surveillance system. METHODS: We compared cases of paralytic poliomyelitis reported to the national surveillance system (maintained by the Centers for Disease Control and Prevention, Atlanta, Ga) with cases recommended for compensation by the National Vaccine Injury Compensation Program, Rockville, Md, and we calculated the observed completeness of reporting to the national system for 1980 through 1991. A capture-recapture method was also used to estimate completeness of reporting, ie, to account for cases potentially missed by both systems. In addition, we reviewed the epidemiology and updated the risk of VAPP based on the most current information on cases of VAPP. RESULTS: From 1980 through 1991, 105 cases of paralytic poliomyelitis were identified by the Centers for Disease Control and Prevention and National Vaccine Injury Compensation Program systems, 98 (93%) of which were VAPP (average, 8.2 cases per year). The observed completeness of reporting to the Centers for Disease Control and Prevention was 94%, and the estimated completeness of reporting (capture-recapture method) was 81%. The overall risk of VAPP was one case per 2.5 million doses of oral poliovirus vaccine distributed. In the sensitivity analysis, the risk estimates of VAPP remained relatively stable throughout a wide range of assumptions regarding underreporting and specificity of the case definition for paralytic poliomyelitis. CONCLUSION: The risk of VAPP remains virtually unchanged from previous estimates despite the inclusion of previously unidentified VAPP cases. Despite the potential for both underreporting and misclassification of cases, our risk estimates were relatively insensitive to either of these biases. Since both of these biases were in opposite directions, and both probably occurred with low frequency, the risk estimates provided in this report appear valid and approximate the "true" risk of VAPP in the United States.