This study compares directives in three different language corpora collected under different conditions: (1) spontaneous spoken data (taken from the British component of the International Corpus of English); (2) spontaneous written data (viz. business letters), and (3) elicited written data (collected through Discourse Completion Tasks). It is shown that there are significant differences between spontaneous and elicited data sets as well as between spoken and written natural data. These differences occur both in the so-called directive head act as well as in the modification strategies accompanying the head act (downgrading and upgrading), resulting in various levels of directness in the realization of directives in all three data sets. These results show the importance of quantitative comparative research not just across data collection methods, but also across discourse genres, based on corpora of authentic speech.