This paper introduces SimClass, a multi-agent classroom simulation framework that leverages large language models (LLMs) to simulate real classroom interactions with user participation. The framework recognizes representative class roles and introduces a novel class control mechanism to enable automatic classroom teaching. Two real-world courses were conducted to evaluate the effectiveness of SimClass. Using the Flanders Interactive Analysis System and Community of Inquiry theoretical framework, the study demonstrates that LLMs can effectively simulate traditional classroom interaction patterns while enhancing user experience. The framework also observes emergent group behaviors among agents, where agents collaborate to create engaging interactions in classrooms to improve user learning processes. SimClass is designed to address the challenges of simulating real classroom environments using LLMs. The framework includes teaching agents and classmate agents, each with specific roles and functions. Teaching agents are responsible for delivering content and managing classroom discussions, while classmate agents engage in peer-to-peer interactions to enhance the learning experience. The framework also includes a Session Controller that manages the flow of interactions based on the current classroom state. The experiments conducted with 48 university students showed that SimClass effectively simulates traditional classroom interactions. The results indicate that the presence of classmate agents enhances user experience in terms of cognitive presence and social presence. The study also found that the classroom environment in SimClass is similar to traditional classrooms in terms of interaction patterns and user experience. The study highlights the potential of LLM-empowered multi-agent systems in virtual classroom teaching. The framework demonstrates the ability to simulate real classroom environments and improve user learning experiences. The results suggest that the use of LLMs in educational settings can lead to more engaging and effective learning experiences. The study also identifies the need for further research to explore the potential of LLMs in educational settings and to improve the effectiveness of multi-agent systems in simulating real classroom environments.This paper introduces SimClass, a multi-agent classroom simulation framework that leverages large language models (LLMs) to simulate real classroom interactions with user participation. The framework recognizes representative class roles and introduces a novel class control mechanism to enable automatic classroom teaching. Two real-world courses were conducted to evaluate the effectiveness of SimClass. Using the Flanders Interactive Analysis System and Community of Inquiry theoretical framework, the study demonstrates that LLMs can effectively simulate traditional classroom interaction patterns while enhancing user experience. The framework also observes emergent group behaviors among agents, where agents collaborate to create engaging interactions in classrooms to improve user learning processes. SimClass is designed to address the challenges of simulating real classroom environments using LLMs. The framework includes teaching agents and classmate agents, each with specific roles and functions. Teaching agents are responsible for delivering content and managing classroom discussions, while classmate agents engage in peer-to-peer interactions to enhance the learning experience. The framework also includes a Session Controller that manages the flow of interactions based on the current classroom state. The experiments conducted with 48 university students showed that SimClass effectively simulates traditional classroom interactions. The results indicate that the presence of classmate agents enhances user experience in terms of cognitive presence and social presence. The study also found that the classroom environment in SimClass is similar to traditional classrooms in terms of interaction patterns and user experience. The study highlights the potential of LLM-empowered multi-agent systems in virtual classroom teaching. The framework demonstrates the ability to simulate real classroom environments and improve user learning experiences. The results suggest that the use of LLMs in educational settings can lead to more engaging and effective learning experiences. The study also identifies the need for further research to explore the potential of LLMs in educational settings and to improve the effectiveness of multi-agent systems in simulating real classroom environments.