Mobile context-aware computing aims at providing services that are optimally adapted to the situation in which a given human actor is. An open problem is that not all mobile services need contextual information at the same level of abstraction, or care for all aspects of the user’s situation. It is therefore impossible to create a unique context model that is useful and valid for all possible mobile services. In this paper we present a compromise: a three-tiered context modeling architecture that offers high-level mobile services a certain freedom in choosing what contextual parameters they are interested in, and on what abstraction level. We believe the proposal offers context modeling power to a wide range of high-level mobile services, thus eliminating the need for each service to maintain complete context models (which would result in severe modeling redundancy if many services run in parallell). Each mobile service must only maintain those parts of the context model that are application-dependent and specific to the mobile service in question. We exemplify the use of the context model by discussing its application to a mobile learning system.