It is commonly believed that agents constitute a single primitive category, distinct from other categories such as patient. This paper challenges such an assumption, arriving at three conclusions: (a) A variety of rules classify agents into affected vs. non-affected types. (b) These rules group affected agents with other affected roles such as patient, dative, and experiencer. (c) The affected/non-affected distinction is more basic than role distinctions such as agent and patient. Among the correlates claimed here for the affected/non-affected contrast is ‘causee’ case-marking, an area that has attracted recent controversy in Comrie 1976 and Cole 1976.