Different kinematic models have been proposed for the triple junction between the North American, Cocos and Caribbean plates. The two most commonly accepted hypotheses on its driving mechanism are (a) the North American drag of the forearc and (b) the Cocos Ridge subduction push. We present an updated GPS velocity field which is analyzed together with earthquake focal mechanisms and regional relief. The two hypotheses have been used to make kinematic predictions that are tested against the available data. An obliquity analysis is also presented to discuss the potential role of slip partitioning as driving mechanism. The North American drag model presents a better fit to the observations, although the Cocos Ridge push model explains the data in Costa Rica and Southern Nicaragua. Both mechanisms must be active, being the driving of the Central American forearc towards the NW analogous to a push-pull train. The forearc sliver moves towards the west-northwest at a rate of 12-14 mm/yr, being pinned to the North American plate in Chiapas and western Guatemala, where the strike-slip motion on the volcanic arc must be very small.