To evaluate the reproducibility of exercise-induced ventricular arrhythmia, 155 men with a mean age of 53 +/- 8 years underwent serial exercise testing 3 to 52 weeks after myocardial infarction. The reproducibility of categorical test responses, that is, the presence or absence of ventricular arrhythmia, was evaluated with the kappa coefficient, which considers negative as well as possible test responses and expresses reproducibility above the chance level. Reproducibility was highest at an intertest interval of 1 to 5 days and was not enhanced by further categorizing premature ventricular complexes as simple or complex based on their frequency or configuration. Continuous response measures such as frequency of premature ventricular complexes yielded higher reproducibility than categorical responses. Continuous response measures appear preferable to categorical responses for evaluating the clinical significance and response to antiarrhythmic therapy of ventricular arrhythmias.