The Antarctic Oscillation (AAO) is the dominant mode of atmospheric circulation variability over the southern hemisphere. It could not only play important roles in climate changes over southern hemisphere, but also exert lots of influences in some regions in northern hemisphere. Due to the lack of widespread instrumental records during historical periods, the understanding of the natural variability of AAO is limited. The purpose of this paper is to reconstruct an austral summer Antarctic Oscillation index (DJF-AAO) focusing on interannual-decadal variability since 1500 AD based on multiple proxies, such as tree-rings, corals, and ice-cores. A Marshall-AAO index derived from 12-station sea level pressure records since 1957 are selected as observational series for calibration. There are 263 variables retained after a series of screening criteria for proxies, to refine the major signatures contained in the proxies by applying principal component analysis, and then a series of screening criteria implemented again for the time coefficient (PC) corresponding to each eigenvector. After that, by applying multivariate regression method the observational AAO-PC relations were calibrated and cross-validated based on the period of 1957-1989, then regressions were employed to compute the DJF-AAO index in 1500-1956. In verification procedure we checked the explained variance (r2), reduction of error (RE), and the standard error (SE). The cross-validation was performed by applying a leave-one-out validation method. During the reconstruction period of 1500-1956, the mean of r2, RE, and SE are 59.9% , 0.47 and 0.67, respectively. These statistical data indicate that DJF-AAO reconstruction is relatively reasonable for the last 460 years approximately. The reconstruction is compared favorably with several existing shorter AAO indexes derived from station SLP records both on the interannual and decadal time scales. The leading periods of the DJF-AAO index are ~2.4, ~2.6, ~6.3, ~24.1, ~37.6 years during the last 500 years, which are all significant at the 95% level.