An algorithm for volume rendering in an environment of parallel virtual machines is presented in this paper. In order to reduce the communication cost, as well as to guarantee the locality of all subtasks, the volume data are divided and or ganized as a series of slices. By maintaining and employing a database of the pe rformance index, the task subdivision algorithm produces an acceptable load bala ncing. An asynchronous binary method is introduced, which merges all partial ima ges in O(logn) time. An efficient developing platform based on the TCP/I P and Socket standards is built. It helps researchers to parallelize various ren dering algorithms on a virtual machine. The algorithm introduced in this paper h as been implemented on this platform, exploiting the classical client/server par adigm. The scalabilities of both the task size and the number of hosts are teste d. The experimental results are demonstrated and analyzed.