To restore confidence in polling, pollsters must recognize the limitations of traditional random sampling and improve methods for diagnosing nonresponse bias.