CAS3D-2, a new three-dimensional (3-D) dislocation model, is developed to model interseismic deformation rates at the Cascadia subduction zone. The model is considered a snapshot description of the deformation field that changes with time. The effect of northward secular motion of the central and southern Cascadia forearc sliver is subtracted to obtain the effective convergence between the subducting plate and the forearc. Horizontal deformation data, including strain rates and surface velocities from Global Positioning System (GPS) measurements, provide primary geodetic constraints, but uplift rate data from tide gauges and leveling also provide important validations for the model. A locked zone, based on the results of previous thermal models constrained by heat flow observations, is located entirely offshore beneath the continental slope. Similar to previous dislocation models, an effective zone of downdip transition from locking to full slip is used, but the slip deficit rate is assumed to decrease exponentially with downdip distance. The exponential function resolves the problem of overpredicting coastal GPS velocities and underpredicting inland velocities by previous models that used a linear downdip transition. A wide effective transition zone (ETZ) partially accounts for stress relaxation in the mantle wedge that cannot be simulated by the elastic model. The pattern of coseismic deformation is expected to be different from that of interseismic deformation at present, 300 years after the last great subduction earthquake. The downdip transition from full rupture to no slip should take place over a much narrower zone.