The establishment of dominant tree regeneration is currently a focus for concern regarding the sustainability of assisted natural regeneration in natural forests (by conservation of natural growth) on Loess Plateau of China. Chinese pine (Pinus tabulaefomis) and Liaodong oak (Quercus wutaishanica) dominate the forests in this region. Results from silvicultural treatments used to regenerate these forests, such as near-natural forest management and artificial promotion of the natural regeneration, may or may not mimic natural forest composition and structure (forest naturalness). The quantitative methods of analysis of forest naturalness are not well known, although previous research addressed forest naturalness either qualitatively or theoretically, and did not consider successional stage. We used gray relational analysis and developed a model to assess forest naturalness by successional stage of five forest types: Liaodong oak, Chinese pine, Oriental arborvitae (Platycladus orientalis), Asian white birch (Betula platyphylla), and Korean aspen (Populus davidiana). Plant composition of each forest community in the four succession stages (forest formation, qualitative growth, competitive selection and climax community) was obtained via cluster analysis. The Climax Adaptation Value (CAV), which represents stage of development, was assigned for each successional stage, and the relative Important Value (IV') of every species was calculated separately. Complex Index (CI) of community was calculated following: CI=∑IV'×CAV. Next, using Analytical Hierarchy Process (AHP), we determined a community characteristic index weight (W_i), and applied gray relational analysis to determine the gray relational coefficient and the gray relational grade (R'_i) among forest type communities. Complex Index of each forest type was one of the most important factors related to forest naturalness. Complex Index ranged from a high of 8.37 for Liaodong oak, to a low of 6.87 for Asian white birch. Chinese pine and Liaodong oak forests had the highest gray relational grade (R'_i= 1.14 and 1.09, respectively) and were at the most stable successional stage, while Korean aspen forests (R'_i=0.79) were the least stable community in the study region. Our results suggest that gray relational analysis is a valuable tool and an effective approach for quantifying forest naturalness, and the results from such analysis can provide insights into vegetation structural and compositional changes and guidance for near-natural forest management and stand improvement. This method of gray relational analysis may be applied to other forest types in the region.