An approach is disclosed for metering usage of cloud computing services at a feature level. In one embodiment, a metering application receives feature status information indicating which features of a cloud computing service are enabled through an event-driven process, in which events are generated and reported to the metering application when features are enabled or disabled, as well as a polling process, in which the metering application periodically queries the statuses of features. Feature status information gathered by the metering application is persisted in a database as collective feature statuses, with each collective feature status being associated with an interval of time during which feature statuses are unchanged. The feature status information obtained and persisted in the database may then be used to, for example, determine licensing fees based on feature usage or report feature usage to facilitate quantitative studies of the usefulness of features.