To address the contradiction between improving compression ratio and reducing metadata cost, a deduplication method based on pre-chunking and sliding window is proposed. A universal performance-analyzing model is also given. In this method, the data objects are pre-chunked based on content, then different chunking strategies are used on the data changing regions and the non-changing regions respectively. A satisfying compression ratio and lower metadata cost can be achieved with a relatively larger expected chunk size. The experimental results on real data indicate that the average compression ratio of the method is higher than the current optimal value and the average time cost is reduced significantly.