Continuous GPS measurements from Colima, Mexico during 4/93--6/01, bracketing the Oct. 9, 1995 M = 8.0 Colima-Jalisco earthquake, provide new constraints on Rivera plate subduction mechanics. Modeling of margin-normal strain accumulation before the earthquake suggests the Rivera-North America subduction interface was fully locked. Transient postseismic motion from 10/95--6/97 is well fit by a model that includes logarithmically-decaying fault afterslip, elastic strain from shallow fault relocking, and possibly a minor viscoelastic response, but is fit poorly by models that assume a dominant Maxwell viscoelastic response of the lower crust and upper mantle, independent of the assumed viscosities. Landward, margin-normal motion since mid-1997 is parallel to but ~75% slower than the pre-seismic velocity. Afterslip alone fails to account for this slowdown. The viscoelastic response predicted by a FEM correctly resolves the remaining velocity difference within the uncertainties. Both processes thus offset elastic strain accumulating from the relocked subduction interface.