The New Madrid seismic zone (NMSZ), a region within the eastern portion of the North American plate where great earthquakes occurred in 1811--1812, provides a type example of large earthquakes within a relatively stable continental interior. Assessing the scale and causes of deformation in such regions, as well estimation of the recurrence interval for the large earthquakes within them, has long been recognized as an important but challenging issue, for which the advent of space-based geodesy provides an important new tool. Here we report initial results from two Global Positioning System (GPS) surveys of a geodetic network spanning the New Madrid seismic zone. Because the expected strain signal is small, the data were analyzed in three different ways. First, site velocities were estimated and analyzed for spatial gradients parallel and perpendicular to the trend of the major strike-slip faults inferred from seismicity. This analysis suggests, but does not require, a few millimeters per year of right-lateral shear across the southern portion of the seismic zone, consistent with the earthquake focal mechanisms. Second, we divided the network into near-field subnetworks and found possibly significant shear associated with approximately E-W shortening within the southern NMSZ and possible N-S shortening across the left step connecting the presumed strike-slip faults. Third, we compared the strain observed along individual baselines to that predicted from the results of earlier near-field surveys which combined GPS and triangulation data. In the southern part of the seismic zone, we found general agreement with the earlier study, although our data suggest lower strain rates. Hence after only 2 years of measurements, the GPS data suggest but do not yet require detectable strain accumulation. However, given that great earthquakes occur in the area and that our results are reasonably consistent with the results of other geodetic studies using different data and with earthquake mechanisms, it seems likely that we are seeing strain accumulation in the southern NMSZ. A simple interpretation of the data as showing about 3 to 5 mm yr-1 of right-lateral fault-parallel motion implies a 1000 to 3300-year recurrence for great earthquakes with 5 to 10-m slip, assuming all the accumulating strain is released seismically. This recurrence interval may be an underestimate, if the observed strains are largely postseismic effects of the 1811--1812 earthquakes, a possibility suggested by the observation that plate-wide data show less motion across the NMSZ. Future surveys should resolve this issue as the longer measurement spans yield more precise velocity estimates. ¿ 1998 American Geophysical Union