In recent years, ecological engineering has become increasingly important in slope protection under climate change. Plants roots can enhance shallow soil strength, reducing the probability of shallow landslides. Predicting the shear strength of root-reinforced soil is crucial for slope stability analysis. Most existing models, which were based on Wu model, assumed roots failed in tension. However, experiments show that roots are often pulled out rather than broken in shear tests, causing the overestimation of shear strength. This paper proposed a modified shear strength prediction model based on the pullout phenomenon of roots during shearing. The model defined total strength of root reinforced soil as the sum of inherent soil strength and additional resistance from root-soil friction, also considering soil suction effect. Direct shear box tests were conducted for alfalfa root reinforced Yili and Yangquan loess in China, examining the effects of root area ratio and root diameter. Results showed that alfalfa roots could increase loess shear strength. For a given root diameter, soil shear strength increased with root area ratio. However, at a consistent root area ratio, the increasing of root diameter led to the decrease of soil shear strength. The shear strengths of two root-reinforced loess soils were predicted using four different approaches, including Wu, Ning, energy-based models, and modified model in this paper. Compared to Wu, Ning, and energy-based models, the proposed model yielded superior accuracy with more accessible parameters, particularly demonstrating robust performance for larger-diameter roots.