Three probabilistic methods for the estimation of seismic risk have been used in Canada. A reevaluation of the extreme value method shows no advantages over the average value method of Milne and Davenport. Conceptual improvements in the underlying assumptions of the latter method are a constrained release of historical earthquakes from their presumed epicenters and the averaging of earthquake rates over variable periods. Risk estimation can then proceed as suggested by Cornell. Comparison of the results of this modification of the average number method shows similar results as the Milne and Davenport average value method. The stability of risk estimates against new earthquakes is improved, but sensitivities at typical sites toward unavoidable deterministic elements in the model are similar to the older method. For certain site-source-seismicity combinations probabilistic estimates of ground motion could become almost quasi-deterministic.