The aim of this study was to determine which, and how many, data items are required to construct a decision support algorithm for early diagnosis of acute myocardial infarction using clinical and electrocardiographic data available at presentation. Logistic regression models were derived using data items from 600 consecutive patients at one centre (Edinburgh), then tested prospectively on 510 cases from the same centre and 662 consecutive cases from another centre (Sheffield). Although performance of the models increased with progressive addition of data inputs when applied to training data, a simple six-factor model was the most effective on test data, yielding accuracies of 84.3 and 83.6% on the two test sets. A model constructed solely of electrocardiographic data performed nearly as well as those incorporating clinical data. Previously published logistic regression models did not perform so well as the models derived from data collected for this study.