ABSTRACT Today many open sources of information are available on the Internet that provide sharing and reusing of learning materials to reduce the cost of designing new courses, save the time of rewriting, and avoid the duplication of efforts. Accordingly, instructors compose their courses or lectures from pieces collected from several sources. Those pieces/chunks are referred to as Learning Objects (LOs) that may be drawn from different Learning Objects Repositories (LORs). However, instructors usually have their own objectives that might differ than those of the original author of the learning materials. In addition, students might differ in skills, backgrounds, and learning styles. All these argue for the instructors' urgent need to have a powerful query mechanism that helps to accurately specify the LOs to be fetched from repositories. In this research, mechanisms that support instructors and e-tutors in selecting the most appropriate learning materials for more effective learning outcomes are investigated. On one hand, instructors need to prepare course materials that meet specific goals such as course objectives and syllabus. On the other hand, students need to have studying materials that match their learning styles and that are built based on their background knowledge. Therefore, the objective of the research is to build a model and an architecture for a Smart e-Learning Assistant (SeLA) that provides instructors and e-tutors with smart assistance in selecting the most appropriate LOs for both Adaptive Course Preparation and Delivery from a higher level perspective. SeLA intelligently rewrites the course objectives according to instructional design theory and then adaptively selects the most appropriate LOs from LOR to align course objectives with students' models. SeLA supports selecting the most appropriate LOs at both stages of the educational process: authoring and delivery. SeLA employs two main theories in building its model: the Revised Bloom's Taxonomy of instructional design (RBT) and Felder-Silverman Learning Style Model (FSLSM). Under this research, a prototype in .NET environment has been developed and evaluated.