This paper presents a brief overview of three interconnected aspects of the quest for universals of translation: (1) the level where translation universals could be located, (2) the possible format of those universals, when formulated in language, and (3) the way to expose them, in view of this level and that format. It is claimed that universals should not be sought on either too low or too high a level, that their format should be probabilistic and conditioned rather than deterministic, and that they should be found and formulated by means of a combination of ‘top-down’ and ‘bottom-up’ research procedures. In a final isntance, the question is tackled of whether the probabilistic formulations themselves (i.e. beyond the mere principle of being probable) would constitute universals, or whether they represent a challenge to the very concept.