In order to realize full cycle sustainable management of Larix gmelinii natural forests in the Khingan Mountains region, this study is based on the data from fixed survey plots of three different successional stages (Betula forests, Betula-Larix mixed forests, and Larix gmelinii forests) within three 100m×100m natural forest stands in Cuigang Forest Farm of the New Forest Bureau. Statistical models (log-normal model), niche models (broken-stick model, niche preemption model, Zipf model and Zipf-Mandelbrot model), and neutral models (zero-sum multinomial model, Volkov model) were used to simulate the species abundance distribution patterns in the tree layer, regeneration layer and different forest layers divided by the tree layer of natural forests at different successional stages. Further, we applied the chi-square test and the Kolmogorov-Smirnov test to select the best-fitting species abundance distribution models. The results indicate: (1) the Zipf-Mandelbrot model had the best fit in different successional stages natural of Larix gmelinii forest communities in Great Khingan Mountain region. (2) The lognormal model had the better fit for the species abundance distribution in stable successional stages, including Betula forests and Larix gmelinii forests, compared to transitional stages communities of Betula-Larix gmelinii mixed forests. (3) Habitat filtering dominates community assembly in the natural Larix gmelinii forests of the Great Khingan Mountain region, with the weakening of habitat filtering and neutral processes as the community successions continue. (4) The community assembly process differed between the dominant ecological processes in the tree layer and the ecological processes in the single-layer in the community. The habitat filtering process gradually weakens in the tree layer of Larix gmelinii forests, while it strengthens in different forest layers divided by the tree layer and the regeneration layer.