In today’s web, many web services are created and updated on the Internet. In many cases, a single service is not sufficient to respond to the user’s request and often services should be combined through service composition to fulfill business goals. Service discovery and service composition can be highly compatible with context, i.e., according to context information, e.g., location, budget and time, services are chosen and composed. Moreover, we include non-electronic services, e.g., restaurants, movie theaters shopping malls and so on, into service composition. Non-electronic services are rarely considered in existing service composition research, however are frequently used in people’s daily life. In this thesis, we provide an approach for using contexts to discover and compose non-electronic services. We present a new context model which is to make it more suitable for service composition. This model is also able to handle both low level sensor data and high level data in predicated logic. Our service composition algorithm uses soft constraints, dissatisfaction of which causes a penalty instead of the fail of planning. With this feature, the service composition algorithm can give the user several “good enough” solutions, instead of null solution. Additionally, a replanning module is developed to refine the solution according to user’s further adjustments of his or her requirements. As a motivating example, a web based Personal Entertainment Planner system is built.