Large-scale parallel computing is a hot spot of current research. Because most of the scientific calculation is data parallel calculation, people would like to solve real problems by means of data parallelism. As for data parallel calculation, one of the critical factors that influence the running performance is the status of data distribution. In this paper, the authors analyze the data distribution algorithm of a prevalent object-oriented data parallel language——pC++ and point out the weaknesses of it. Then they present an improved algorithm. The results of the experiment show that it is superior to the former algorithm.