Adverse drug events cause a large number of injuries, and adverse events caused by medications administered in the face of known allergies represent an important preventable cause of patient harm. Computerized systems can effectively prevent reactions due to known allergies, but building an effective allergy prevention feature is challenging and presents many interesting informatics issues that have both methodological and operational implications. In this paper, we present the experiences from one large delivery system in delivering allergy-related decision support, discuss some of the different approaches that we have used, and then propose a future approach. We also discuss the methodological, behavioral, and operational issues that have arisen which have a major impact on success. Key factors in drug-allergy checking include storing patient allergy data in a single common repository, representing allergy data using suitable terminologies and creating groups of allergies for inferencing purposes, being judicious about which allergy warnings to display, conveying the reaction that the patient has experienced when exposed to the drug to inform the provider of the importance of the warning, and perhaps most important, implementing strategies to optimize the likelihood that allergy information will be entered.