In the last decade, selection and composition of Web services have drawn increasing attentions. However, real-life services are not only web services but complicated eco-systems composed of various service elements like human, resources, environment, etc, and existing service composition methods cannot be directly applied to such real-life service composition scenarios. In this paper, we propose a conceptual model for generalized real-life service composition. In this model, service behaviors are abstracted as service components and uniformly described by XML. Various service requirements raised by customers are completely listed and classified. Based on these works, a genetic algorithm for real-life service composition is presented to select the best matching service components and compose them together. The algorithm's effectiveness of obtaining optimal solution is proved by a prototype system.