Two contrasting models have been proposed to describe the present-day motions between the Caribbean Plate and neighboring plates. One model, based on both North America-Caribbean and Cocos-Caribbean data, assumes that North America-Caribbean motion is reflected by the spreading rate (approximately 2 cm/yr) inferred from magnetic anomalies at the Cayman Spreading Center and the azimuths of nearby transforms (Jordan, 1975). This geometry was used by Minster and Jordan (1978) in deriving global plate motion model RM2. The other model, based only on North America-Caribbean data, uses rates and azimuths inferred from the geometry of the Lesser Antilles Wadati-Benioff zone (Sykes et al., 1982). The Cayman Spreading Center data were discounted assuming that they underestimate the full North America-Caribbean motion owing to the complex tectonics of the Greater Antilles. The latter model assumes plate convergence at the Lesser Antilles at a rate about twice as fast as, and an azimuth differing by 25¿ from, the Jordan model. The two models also differ in their predictions of South America-Caribbean motion; the Jordan model predicts oblique convergence whereas Sykes et al. predict oblique divergence.

We use the NUVEL-1 global relative motion data set, which incorporates recent data not used in earlier studies, to discriminate between the alternative models. We find that the Jordan geometry provides a better fit to the direction of North America-Caribbean motion. Moreover, we find that models based on this geometry better fit the direction of Cocos-Caribbean motion observed in earthquake slip vectors along the Middle America Trench. Since the prediction of this direction depends strongly on the rate of North America-Caribbean motion, we conclude that the Jordan model better describes Caribbean Plate motions. The discrepancy between the two models results from the different methods of estimating plate motions. Sykes et al. (1982) report that they determined a Caribbean-North America convergence rate of approximately 4 cm/yr from the length of the Lesser Antilles Wadati-Benioff zone; we employ a similar procedure and obtain a rate closer to 2 cm/yr. We also find that the method they report using to determine the convergence direction is unable to distinguish between the alternative azimuths. We thus conclude that estimates of the rate and direction of plate motion from Wadati-Benioff zone configurations are insufficiently robust and that such estimates are better derived using rates from magnetic anomalies and azimuths from transform faults and slip vectors. We further conclude that successful estimates of relative motion require internal consistency between the Euler vectors of all relevant plates, as demonstrated by the powerful Cocos-Caribbean constraint not incorporated in the Sykes et al. model. ¿ American Geophysical Union 1988