Large 3D meshes are emerging as the new media for various applications but are still hard to use in mobile applications, due to the limited resources of mobile systems. This paper introduces a large 3D mesh streaming framework which flexibly deals with the limited resources of mobile systems and also provides a user with interactive controls and random accessibilities. To reduce resource usage, our framework presents a uniform mesh partitioning algorithm, in which each partition of a large mesh has the same number of vertices. Our uniform partitioning is based on a k-d tree clustering and extended for out-of-core meshes. A median search with heaps in the main memory is designed for faster external sorting. Large 3D meshes are transformed into a set of partitioned and simplified meshes in the server. For interactive 3D browsing, on mobile devices, our framework presents a mobile 3D viewer which is hierarchically designed with intuitive interfaces. A user can experience rapid 3D searches with 3D previews of simplified meshes. Multi touch inputs can control zooming and the level of detail in meshes automatically. Double tap touches enable a user to randomly select a region of a large mesh, and a set of partitions for the selected region will be streamed and displayed on a mobile client. As a result, our framework enables a user to browse large 3D meshes on a mobile system interactively, while optimizing system resource usage and protecting the original data in the server.