A new noninvasive method for early detection of ischemic heart disease based on singular value decomposition is introduced. The method automatically explores waveforms of an ECG by making projections of individual ECG leads onto axes formed by the first three components of the singular value decomposition performed for the whole group of ECGs, including cases with and without coronary artery lesions (460 ECGs). Detection of the presence of lesions is performed with 73% sensitivity and 83% specificity. Sensitivity and specificity become lower when specifying which coronary artery contains lesions: 69% and 77%, respectively, for the right coronary artery, 66% and 82%, respectively, for left anterior descending artery. 74% and 71%, respectively, for the left circumflex artery. The method is being developed so that it might be included into the programs of automatic ECG analysis, particularly for population studies.