This paper addresses two points of a research agenda set by Levinson and Wilkins (2006) for the analysis of spatial information processing in natural languages. First, a pragmatic model is proposed that relies mostly on general pragmatic principles to inferentially disambiguate semantically underdetermined spatial expressions. Second, this model is claimed to be cross-linguistically applicable and leads to a universal generalisation about the pragmatics of spatial frames of reference. In the second part, a report is provided of an experimental design in support of the theoretical points raised before. Finally, cross-linguistic evidence from Tzeltal is proposed to back the generalisation argued for. To conclude, the evidence presented is taken to strongly support the predictions made by a pragmatic model of spatial language. Such a model offers a fresh take on the central question of universality vs. Linguistic diversity in spatial language as it uses semantic and pragmatic universals to explain linguistic variation.