To improve the quality and consistency of health care, evidence-based medicine (EBM) was proposed to promote the wide adoption of current best evidence to make decisions about care of individual patients. The practice of EBM in Radiation Oncology is a process of integrating clinical expertise, patient's expectation, and research evidence to support decision-making in Radiation Therapy (RT). One of the goals of the RT decision-making aims to design an ideal RT plan that achieves most damage to target treatment site and least harm to surrounding healthy organs and tissues. This aim requires radiation oncologists to understand and maintain up-to-date Radiation Oncology knowledge, also known as the clinical evidence published in clinical guidelines and clinical research studies, such as radiation-induced adverse events, dosimetric criteria recommendation, and meta-analysis of randomized controlled clinical trials.As the amount of clinical evidence increases, it is becoming increasingly difficult for clinicians to maintain and adopt the best and most up-to-date clinical evidence in their clinical practices. This demands the development of effective systems for automated and intelligent clinical decision support (CDS), which relies on knowledge engineering methods to translate narrative clinical knowledge into computable forms and enable reasoning of the computerized knowledge. In the domain of RT, we believe that computerized Radiation Oncology knowledge will improve the ability and quality of intelligent decision-making in an efficient way.This dissertation aimed to advance the state-of-the-art of research towards evidence-based medicine in general and with a specific focus on intelligent decision support for radiation therapy. First, we explored radiation-induced adverse events and their grading standards used in clinical research studies using ontological modeling and text mining methods. Second, we investigated the challenges and developed a framework for extracting Radiation Oncology knowledge from clinical guidelines and clinical research studies. Third, we focused on the specific and challenging problem of uncertainty nature of human biological systems and biomedical research approaches. Toward this end, we investigated the feasibility of probabilistic models for representing extracted RT knowledge and the ability of performing reasoning. Specifically, we developed novel methods to encode uncertain Radiation Oncology knowledge using Markov Logic Networks and conducted a study of quantifying uncertainties in Radiation Oncology clinical evidence. We demonstrated the feasibility of using the proposed methods as a general knowledge engineering framework for representing complex and uncertain knowledge in Radiation Oncology for decision support.