Multiple dipole models of activation of the heart, in combination with a model of the thorax as an electrical conductor relate the action of individual cells to features of the electrocardiogram. The models can be used to develop diagnostic electrocardiographic criteria for ventricular hypertrophy, myocardial infarction or conduction disorders. Criteria derived with the use of Holt, Barnard and Lynn's model proved to be superior to well-established empirical rules. Their approach restricts position, direction and polarity of the current dipole vectors in the model. In the Selvester model, and later in the Ritsema van Eck model, in addition to these restrictions time-constraints are imposed on the dipole activity. In a validation study of the Ritsema van Eck model activation of a beating dog heart is compared in detail with simulated activation in a computer replica of the same heart. Myocardial spread of activation is found to be stimulated correctly. Simulated spread of activation along the Purkinje fiber network however, differs considerably from reality. In apical regions excitation spreads in reality much faster than in the model, in basal regions much slower. The model can be improved by slowing endocardial excitation from apex to base with the time-course of activation. Further improvement is obtained by the use of a more detailed pattern of endocardial excitation, based on studies of endocardial excitation in opened ventricles. The model can predict the time-course of activation with an accuracy of 5% of total ventricular activation time.