Much research in recent years has focused on question answering. Due to significant advances in answering simple fact-seeking questions, research is moving towards resolving complex questions. An approach adopted by many researchers is to decompose a complex question into a series of fact-seeking questions and reuse techniques developed for answering simple questions. This thesis presents an alternative novel approach to domain-specific complex question answering based on consistently applying a semantic domain model to question and document understanding as well as to answer extraction and generation. This study uses a semantic domain model of clinical medicine to encode (a) a clinician's information need expressed as a question on the one hand and (b) the meaning of scientific publications on the other to yield a common representation. It is hypothesized that this approach will work well for (1) finding documents that contain answers to clinical questions and (2) extracting these answers from the documents. The domain of clinical question answering was selected primarily because of its unparalleled resources that permit providing a proof by construction for this hypothesis. In addition, a working prototype of a clinical question answering system will support research in informed clinical decision making. The proposed methodology is based on the semantic domain model developed within the paradigm of Evidence Based Medicine. Three basic components of this model - the clinical task, a framework for capturing a synopsis of a clinical scenario that generated the question, and strength of evidence presented in an answer - are identified and discussed in detail. Algorithms and methods were developed that combine knowledge-based and statistical techniques to extract the basic components of the domain model from abstracts of biomedical articles. These algorithms serve as a foundation for the prototype end-to-end clinical question answering system that was built and evaluated to test the hypotheses. Evaluation of the system on test collections developed in the course of this work and based on real life clinical questions demonstrates feasibility of complex question answering and high accuracy information retrieval using a semantic domain model.