Early Warning Systems (EWS) are an essential component of effective disaster management. In the past decade we witnessed the evolvement of highly sophisticated warning systems, in particular for large-scale natural disasters such as floods, storms, and tsunamis. These systems increasingly require the use of advanced information processing for accurate monitoring and prediction as well as targeted alerting under near real-time requirements. Yet, a common reference architecture for such EWS is missing that provides a solid basis for the application of advanced information processing methods. In this paper, we propose a general reference architecture that provides a systematic and common understanding of the functional elements of an EWS solution. Beside providing the foundation for better EWS-specific systems engineering, this architecture can serve as a basis for the identification of application potentials, effective design, implementation, deployment, operation, evolution and re-use of computational intelligence modules within an integrated EWS solution.