The role of multifidus in the biomechanics of lumbar spine remained unclear. This study aimed to investigate the role of multifidus in the modeling of lumbar spine and the influence of asymmetric multifidus atrophy on the biomechanics of lumbar spine. This study considered five different multifidus conditions in the trunk musculoskeletal models: group 1 (with entire multifidus), group 2 (without multifidus), group 3 (multifidus with half of maximum isometric force), group 4 (asymmetric multifidus atrophy on L5/S1 level), and group 5 (asymmetric multifidus atrophy on L4/L5 level). In order to test how different multifidus situations would affect the lumbar spine, four trunk flexional angles (0°, 30°, 60°, and 90°) were simulated. The calculation of muscle activation and muscle force was done using static optimization function in OpenSim. Then, joint reaction forces of L5/S1 and L4/L5 levels were calculated and compared among the groups. The models without multifidus had the highest normalized compressive forces on the L4/L5 level in trunk flexion tasks. In extreme cases produced by group 2 models, the normalized compressive forces on L4/L5 level were 444% (30° flexion), 568% (60° flexion), and 576% (90° flexion) of upper body weight, which were 1.82 times, 1.63 times, and 1.13 times as large as the values computed by the corresponding models in group 1. In 90° flexion, the success rate of simulation in group 2 was 49.6%, followed by group 3 (84.4%), group 4 (89.6%), group 5 (92.8%), and group 1 (92.8%). The results demonstrate that incorporating multifidus in the musculoskeletal model is important for increasing the success rate of simulation and decreasing the incidence of overestimation of compressive load on the lumbar spine. Asymmetric multifidus atrophy has negligible effect on the lower lumbar spine in the trunk flexion posture. The results highlighted the fine-tuning ability of multifidus in equilibrating the loads on the lower back and the necessity of incorporating multifidus in trunk musculoskeletal modeling.