In the paper, I distinguish between the semantic and the “direct” approach to event ontology. The first approach, employed by D. Davidson, starts with logical analysis of natural language. This analysis uncovers quantification over the domain of events. Thus, we have ontological commitment to events and, at the same time, also a suggestion of how to view their nature. The second approach, used by J. Kim and D. Lewis, deals with events “directly”, i.e. not by analyzing language first. Events are postulated because they are useful in other theories (of causation, explanation, etc.) and their nature is adjusted to the needs of these theories. In the paper, I analyze both approaches and outline their problems and advantages. I conclude that we should conditionally prefer the latter approach on methodological grounds. This preference is based on the assumption that submitting hypotheses to tests seems to be a crucial part of metaphysical methodology. Since the “direct” approach to event ontology allows for more testing, it should be preferred over the semantic approach.