Heart disease is a notoriously dangerous disease which possibly causing the death. An electrocardiogram (ECG) is used for a diagnosis of the disease. It is often, however, a fault diagnosis by a doctor misleads to inappropriate treatment, which increases a risk of death. This present work implements k-nearest neighbor (K-NN) on ECG data to get a better interpretation which expected to help a decision making in the diagnosis. For experiment, we use an ECG data from MIT BIH and zoom in on classification of three classes; normal, myocardial infarction and others. We use a single decision threshold to evaluate the validity of the experiment. The result shows an accuracy up to 87% with a value of K = 4.