Previous studies on the English conative alternation have resorted to semantic constraints like “attempt”, “motion”, and “contact” on verbs and have taken Levin’s (1993: 41-42) verb classification about the alternation for granted. This study, based on authentic corpus data extracted from BYU-BNC, shows that simple semantic-constraint-based approaches are insufficient and Levin’s (1993) classification is problematic. In particular, this study shows that the verb classes that participate in the conative alternation in English can be classified in terms of entailments about change, motion, and contact and that some classes and subclasses exhibit idiosyncratic properties which cannot be captured by the previous approaches and Levin’s (1993) classification.