This paper compared two multi-document summarization systems we developed. One system used hierarchical sentence clustering algorithm to find the important information, while the other system mainly adopted hierarchical Latent Dirichlet Allocation (hLDA) topic model to obtain the sub-topics of multi-document data. Both of the two systems are evaluated and compared on TAC 2010/TAC 2011 data using the ROUGE testing method with same parameters' setting. The results have shown that the hLDA system has got some improvement compared with the clustering system. And normally in ROUGE testing, results from non-stopwords are better than those from stopwords.