We have obtained three epochs of high-quality dual-frequency GPS data at 15 stations in the Luzon, Philippines, region since the July 16, 1990, Ms 7.8 earthquake, from which we generate horizontal velocity fields at the Earth's surface for the 1993-1996 and 1996-1998 periods. We use these velocities, plus 1996-1998 GPS velocity results reported elsewhere from 13 additional stations, to show that present-day deformation in Luzon is dominated by strike-slip motion along the Philippine fault system. The measured strike-slip rate across central Luzon is faster than the expected long-term slip rate on the fault, which we attribute to postseismic deformation following the 1990 earthquake. The 1993-1996 and 1996-1998 deformation patterns cannot be statistically distinguished within the ~10% uncertainties of the data. If the observed deformation is interpreted using an elastic half-space model with uniform slip below a subsurface locking depth, we find a fault slip rate of 40 mm/yr and a locking depth of ~15 km. This is too fast to be a steady interseismic rate, as it disagrees with what is known from paleoseismic and other evidence about earthquake recurrence and long-term slip rate on the Philippine fault system in central and northern Luzon. Also, a 15 km locking depth for steady interseismic slip is probably too shallow compared to the depth of rupture in the 1990 earthquake. If the deformation is instead interpreted with two types of viscoelastic model, we find that the observed velocities can be fit well for a range of values of lower lithosphere viscosity and long-term slip rate. Using a two dimensional (2-D) viscoelastic coupling model, the minimum allowable lower lithosphere viscosity is 0.5 x 1019 Pa s, with an associated long-term slip rate of 15-22 mm/yr. Faster long-term slip rates also fit well, with correspondingly higher values of viscosity. With a possibly more realistic model that includes 3-D viscoelastic effects we find preferred long-term velocities in the range 20-35 mm/yr and viscosities in the range 2-6 x 1019 Pa s. Our preferred viscosities correspond to Maxwell times on the order of 7-20% of the typical earthquake recurrence interval and are several times higher than values found for a number of Californian strike-slip earthquakes. Other implications of our 1993-1998 surface velocities are that the Philippine Sea-Eurasia (PH-EU) Euler vector of Seno et al. <1993> is a better description of the PH-EU convergence rate than a Euler vector recently estimated from GPS velocities in the northern PH and that the great majority of the normal component of PH-EU plate convergence is taking place west of Luzon, presumably largely at the Manila Trench. We also find that Luzon is rotating counterclockwise relative to PH at 1¿-2¿/Myr, only ~25% of the rate recently suggested by other workers on the basis of limited GPS data.