WebGIS servers send digital maps to users. For each request-response round, the servers access map data in batches. The access has a feature called multiscale, that is, the map scales selected by users determine the map detail levels. The access method based on R-tree is not adaptive to the multiscale and batch feature. It has two problems: (1) the data records of features of the same level are not clustered in disks; (2) the granularity of data I/O is too small. So accessing map data for display is unefficient. This paper presents a novel spatial index called Multilevel R-tree, which can solve the two problems. The statistics from experiments show that for range queries, the access method based on multilevel R-tree is much more efficient than the one based on R-tree and can improve the response speed of WebGIS Servers.