Background: Existing scores for predicting 30-day hospital readmissions have limited practical use given the complexity of required data elements. Objectives: The purpose of this study was to develop a simple predictive tool to identify patients at high risk of readmission. Methods: Medical or surgical patients who were discharged alive from Regions hospital between 7/1/2006 and 6/30/2009 were identified from administrative data. Our primary outcome was 30-day readmission to Regions hospital. We investigated 80 independent variables as predictors of readmission: demographic and utilization data, care characteristics and discharge medications. Logistic regression was used to find odds ratios and discrimination provided by each predictor. Observations were randomly allocated to training or validation datasets. The training data was used to construct several multivariate models of 30-day readmissions, which were selected for predictive ability, parsimony, and clinical ease of use. From these models, we developed an integer scale to indicate a patient's risk of readmission, based on the magnitude of the observed regression coefficients. Results: Our sample consisted of 38,636 patients. The selected multivariate model had six independent predictors of readmission: a count of ICD-9 codes, a count of past year hospitalizations, indication of >2 ED visits in the past year, indication of being in an operating room, indication of either heart failure or COPD, and discharge to a skilled nursing facility. The composite risk score derived from these predictors ranges from 0 to 16, with higher scores indicating greater risk. This rule showed good discrimination (c-statistic =0.670) when applied to the holdout data.