Abstract: There has been a growing interest in studying the unusual signals of drug-event combinations in post-market safety surveillance. We first briefly review most commonly used data mining methods for the analysis of spontaneous adverse event (AE) reports, and then propose a new method based on a Poisson model for drug-AE cell counts. A maximum likelihood ratio test is developed for testing the null hypothesis that the risk from each drug within a class of drugs, and for a particular AE, is same versus the alternative hypothesis that the risk is higher for at least one drug within the class of drugs and for the fixed AE. The performance characteristics such as Type I error, power, and sensitivity of the test are studied through a simulation study. The proposed method is shown to maintain Type I error and has less misspecified signals compared with other competing methods. The method is then applied to Food and Drug Administration (FDA)'s Adverse Event Reporting System (AERS) data with cases occurred from 2004-2008.