On-line identification of soil parameters is a pre-condition of operating performance optimization and control for unmanned ground vehicles (UGV). Inverse calculation from measured vehicular operating parameters is a prevalent methodology. However, it inherently suffers from a multiple-solution problem caused by the coupling of soil parameters in terramechanics equations and an accuracy problem caused by the influences of state noise and measurement noise. These problems in tractive-force-related soil parameters identification were addressed here for air-cushioned vehicles (ACV) by taking advantage of their additional degree of control freedom in vertical force. To be specific, a g-function algorithm was proposed to solve the multiple-solution problem from reproductive tractive force equations; de-noising techniques consisting of mean-effect strategies, sampling points selection and sample rearrangement were employed to solve the accuracy problem. A series of experiments were conducted to evaluate these techniques at different noise levels and in different soil conditions. They got satisfactory results in terms of data utilization ratio, identification accuracy and performance stability. The contribution of the paper lies in inventing a novel algorithm for unique and accurate identification of tractive-force-related soil parameters without making any simplification to the original terramechanics equation and with robustness to variations of noise level and soil condition.