The semantics of context are examined, considering concepts of relevance, situations, and relationships. We define a situation as a set of relationships and a context as a situation that a) provides expectations for constituent entity states or b) is deemed relevant to the solution of an inference or response problem. The use of context variables in inferencing is examined. Predictive models as used in inferencing are construed as estimates of state distributions. The uses of context in inferencing can be differentiated into categories of target and information source characterization methods, appropriate to different assumptions concerning the quality of available prior models and observational data.