Three multivariate methods for predicting death within 1 year for patients discharged after acute myocardial infarction were evaluated: Cox model, discriminant function analysis and recursive partitioning. Discriminant function analysis was utilized to predict a new myocardial infarction (any new or nonfatal infarction). A Cox classification model developed in a population of 260 patients (group 1) discharged after myocardial infarction was tested in 886 patients from the same institution (group 2) and 582 patients from another institution (group 3). Discriminant function analysis and recursive partitioning were developed in group 2 and tested in group 3. Data gathered during the entire period of hospitalization were utilized. The important variables (ordered as selected by the analyses) for the end point death were: heart failure, ventricular tachycardia and atrioventricular block in the Cox model and heart failure, previous myocardial infarction, age and ventricular premature beats in the discriminant function analysis. For the end point new myocardial infarction, the important variables were: previous myocardial infarction, heart failure, extension of infarction during the acute phase and infarct site. For predicting death and survival within 1 year, each of the three schemes was comparable. For estimating the actual risk of death, the Cox model was best. Recursive partitioning had the advantage of using only one variable--heart failure. Total correct classification ranged from 65.4 (Cox model) to 71.6% (discriminant function analysis) for the original population (groups 1 and 2) and from 47.9 (discriminant function analysis) to 54.3% (recursive partioning) when the schemes were tested in patients in group 3. The Cox model and discriminant function analysis were able to correctly predict over half of the new infarctions within 1 year.