The effect of a drug on an adverse outcome may be difficult to detect if the adverse outcome is rare or has a high background rate in the study population. Clinical trials are typically not powered to test for adverse outcome effects. Often the specific safety outcomes are not known prior to the trials and information may not be collected to measure the outcomes accurately. This talk discusses some lessons in the post-market analysis of safety signals based on two recent projects in the Quantitative Safety and Pharmacoepidemiologic Group of CDER/FDA. The first example is meta-analysis of trial data on the association of suicidality and antidepressants. The second example is propensity score analysis of an observational study of adverse outcomes associated with a cardiac-surgery anti-bleeding agent.