The breeding program for Acacia mangium has entered advanced-generation breeding cycles through adopting a recurrent selection system and a sub-lining breeding population. Genetic variation changes along the successive generations could affect its genetic stability on wide ranges of sites. The aim of this study is to observe genetic stability in third-generation progeny trials of A. mangium established at three different sites in Indonesia. Analysis was conducted, including single-site and multi-sites analyses for height, diameter, and stem forking that were grouped into two sets of analysis based on the genetic background of the trial: SET01 for the single sub-line and SET02 for the composite sub-lines. Index selection for multiple-traits was then used to identify the family changing ranks for multiple-traits and genetic gain prediction. The results showed that the recurrent selection system adopted in the breeding strategy for single-site analysis could maintain sufficient genetic variance of A. mangium in the third-generation progeny trial. Family heritability was moderate to high for almost all traits. However, a strong genetic-environment interaction (G × E) exists in multi-sites analysis for the single sub-line population (SET01), indicating a less sufficient genetic variation and a low Type B genetic correlation in anticipating a wider range of environment. On the contrary, compositing selected family from several sub-lines (SET02) could diminish the strength of G × E and increase Type B correlation. Selection and genetic gain prediction could be more effective in multi-sites analysis for SET02, but it was less effective for SET01. The results imply that adopting a recurrent selection system in advanced-generation breeding of A. mangium should consider structuring the breeding population. It could be practiced by compositing selected superior families from several sub-lines into one breeding population to maintain high genetic stability, while increasing genetic diversity and productivity.