Considering the fuel consumption and soil compaction, optimization of the performance of tractors is crucial for modern agricultural practices. The tractive performance is influenced by many factors, making it difficult to be modeled. In this work, the traction force and tractive efficiency of a low-power tractor, as affected by soil coefficient, vertical load, horizontal deformation, soil compaction, and soil moisture, were studied. The optimal work of a tractor is a compromise between the maximum traction force and the maximum tractive efficiency. Optimizing these factors is complex and requires accurate models. To this end, the performances of soft computing approaches, including neural networks, genetic algorithms, and adaptive network fuzzy inference system, were evaluated. The optimal performance was realized by neural networks trained by backpropagation as well as backpropagation combined with a genetic algorithm, with a coefficient of determination of 0.955 for the traction force and 0.954 for the tractive efficiency. Based on models with the best accuracy, a sensitivity analysis was performed. The results showed that the traction performance is mainly influenced by the soil type; nevertheless, the vertical load and soil moisture also exhibited a relatively strong influence.