A data set comprising 110 spreading rates, 78 transform fault azimuths, and 142 earthquake slip vectors has been inverted to yield a new instanteous plate motion model designated Relative Motion 2 (RM2). The model represents a considerable improvement over our previous estimate, RM1 . The mean averaging interval for the spreading rate data has been reduced to less than 3 m.y. A detailed comparison of RM2 with angular velocity vectors which best fit the data along individual plate boundaries indicates that RM2 performs close to optimally in most regions, with several notable exceptions. The model systematically misfits data along the India-Antarctica and Pacific-India plate boundaries. We hypothesize that these discrepancies are manifestations of internal deformation within the Indian plate; the data are compatible with northwest-southeast compression across the Ninetyeast Ridge at a rate of about 1 cm/yr. RM2 also fails to satisfy the east-west trending transform fault azimuths observed in the French-American Mid-Ocean Undersea Study area, which is shown to be a consequence of closure constraints about the Azores triple junction. Slow movement between North and South America is required by the data set, although the angular velocity vector describing this motion remains poorly constrained. The existence of a Bering plate, postulated in our previous study, is not necessary if we accept the proposal of Engdahl and others that the Aleutian slip vector data are biased by slab effects. Absolute motion models are derived from several kinematical hypotheses and compared with the data from hot spot traces younger than 10 m.y. Although some of the models are inconsistent with the Wilson-Morgan hypothesis, the overall resolving power of the hot spot data is poor, and the directions of absolute motion for the several slower-moving plates are not usefully constrained.