[1] Probabilistic seismic hazard analyses (PSHA) are commonly based on frequencymagnitude statistics from 50-100 yearlong earthquake catalogs, assuming that these statistics are representative of the longer-term frequency of large earthquakes. We test an alternative PSHA approach in continental western Canada, including adjacent areas of northwestern U.S.A., using regional strain rates derived from 179 Global Positioning System (GPS) horizontal velocities. GPS strain rates are converted to earthquake statistics, seismic moment rates, and ground shaking probabilities in seismic source zones using a logic-tree method for uncertainty propagation. Median GPS-based moment rates and shaking estimates agree well with those derived from earthquake catalogs in only two zones (Puget Sound and mid-Vancouver Island). In most other zones, median GPS-based moment rates are 6-150 times larger than those derived from earthquake catalogs (shaking estimates 2-5 times larger), although the GPS-based and catalog estimates commonly agree within their 67% uncertainties. This discrepancy may represent an under-sampling of long-term moment rates and shaking by earthquake catalogs in some zones; however a systematic under-sampling is unlikely over our entire study area. Although not demonstrated with a high confidence level, long-term regional aseismic deformation may account for a significant part of the GPS/catalog discrepancy and, in some areas, represent as much as 90% of the total deformation budget. In order to integrate GPS strain rates in PSHA models, seismic versus aseismic partitioning of long-term deformation needs to be quantified and understood in terms of the underlying mechanical processes.