Abstract Languages differ widely from one another in the extent to which they are transparent, i.e. obey one-to-one relationships between meaning and form. Transparency, in turn, is an important factor in the learnability of languages. This paper first sets out a framework for the study of transparency and subsequently studies cross-linguistic differences in transparency, using the theory of Functional Discourse Grammar as its point of departure. Transparent and non-transparent features of languages are systematically defined using the multi-level architecture of this model of language, representing them as mappings between and within levels. In applying this framework to a sample of 30 languages it is shown that the (non-)transparent features investigated can be ordered into an implicational transparency hierarchy, and that as a result the languages of the sample can be ranked in terms of their degrees of transparency as well. Finally, the consequences of these findings for the learnability of languages are discussed.