Homogenization of climate observations remains a challenge to climate change research. To examine the influence of metadata on homogenizing climate data, the authors applied many test methods including SHNT (standard normal homogeneity test), Buishand test, Pettitt test, MASH (multiple analysis of series for homogenization), TPR (two phase regression), and Von Neumann ratio test to the Sichuan annual temperature series during 1960-2009. The results show that there are 42 stations existing discontinuity, about 40%; in the inhomogeneous stations, 29 were caused by station migration, and 19 caused by the replacement of equipment. The heterogeneity of the station was increasing during 1960s-1970s, 1980s-1990s and after 2000, and most breakpoints were caused by migration. Analyzing and assessing the sensitivity and applicability of the test methods show that, 48.6% of the SNHT test results were consistent with Buishand and Pettitt, and TPR method which does not involve adjacent stations was consistent with other methods in a low rate. In reference to the topography of Sichuan, it is found that SNHT makes the breakpoint being revealed with the highest rate in the basin and mountain stations, reaching 67.6% and 57.1%, and the missed rate and false positives were low. SNHT and Buishand methods make the breakpoint being revealed with the highest rate in plateau stations, and the missed rate and false positives were low with Buishand. Therefore, in view of reducing dependence on metadata, the homogenization test with the integrated application of variety test methods should contribute to climate change research effectively.