Innovations in network and information technology have transformed traditional classroom lectures into new approaches that have given universities the opportunity to create a virtual laboratory. However, there is no systematic framework in existing approaches for the development of virtual laboratories. Further, developing a virtual laboratory from scratch is time consuming and costly. This article proposes a systematic framework to classify the activities between learners and instructors in the laboratory and to design the mobile agent-based virtual laboratory by wrapping the existing CAI tools without knowing the source code. Using the existing CAI tools can reduce the time and cost in constructing a virtual laboratory. The framework consists of three parts: mobile agent execution environment, mobile agent and learning platform. Moreover, various mobile agent design patterns are provided for users to design and implement virtual laboratories. This framework of patterns could make mobile agent based virtual laboratories easier to design and understand. The authors' framework has the following merits: adaptability, cost-effectiveness and collaboration. Their framework has demonstrated its feasibility in several applications, including digital circuit, language learning and digital signal processing laboratories. (Contains 1 table and 14 figures.)