Sperber and Wilson's relevance theory draws a distinction between conceptual and procedural encoding. This is a cognilive distinction, according to which conceptual encoding contributes to the construction of conceptual representations and procedural encoding constrains the inferential computations performed over these. Recent work within the relevance theory framework (e.g., Haegeman 1989, Klinge 1993, Nicolle 1997a has characterized grammatical markers of modality, tense, and aspect äs exponents of procedural encoding. Although these accounts are synchronic, the general claim ihat grammatical markers encode procedural Information should also be compatible with evidence concerning the historical development of grammatical markers (a process known äs grammaticalization). In this article, I demonstrate that a procedural characterization of grammatical markers is not only compatible with research into grammaticalization, but also sheds light on the following problematic issues. First is the question of whether grammaticalization is semantically gradual (äs is generally assumed) or semantically instantaneous äs Givon (1991) claims; I provide evidence in favor of Givon s view. Secondly, a procedural analysis suggests an answer to the question of what semantic mechanism initiates grammaticalization. Finally I demonstrate how the distinction between conceptual and procedural encoding accountsfor the phenomenon of semantic retention (residual lexical meaning) in grammatical markers.