<h3>Purpose/Objective(s)</h3> Studies have shown that head and neck tumors are impossible to achieve an ideal match over the entire length of the neck due to the wide range of radiotherapy target areas. Previous studies have shown random deformation errors of 0.5 to 3.6 mm in the head and neck, the difference between any two different matched areas can reach 2 to 6 mm. we aimed to evaluate the role of neck muscle group training on the repeatability of the pose in patients treated with radiotherapy for head and neck tumors. <h3>Materials/Methods</h3> Patients treated with radiotherapy for head and neck tumors were randomly assigned (1:1) to a study group and a control group. The study group started cervical spine exercises+massage training one week before the start of radiotherapy until the end of radiotherapy. In the control group, conventional treatment without specific interventions was performed. The primary endpoint was the difference in the reproducibility of cervical spine set-up between the two groups, and the secondary endpoints were the differences in cervical curvature changes on lateral cervical radiographs before and after treatment between the two groups, and the differences in the timing and extent of side effects before, during, and after treatment between the two groups. <h3>Results</h3> A total of 947 CBCT scans were performed in 166 patients, including 479 CBCTs in 83 patients in the study group and 468 CBCTs in 83 patients in the control group. 26.51% (22/83) of the study group and 33.73% (28/83) of the control group had an error of ≥3 mm during the first treatment, (P=0.31). Patients with clivus and neck error >3mm in anterior-posterior direction during treatment accounted for 8.43% (7/83) of the study group and 26.51% (22/83) of the control group(P=0.002). The number of repeated setups except the first treatment was 2.42% in the study group and 7.02% in the control group (P=0.01). The error range from clivus to neck was (7, 3, 7) mm. setup error: in the clivus position of the two groups in the X, Y, Z three directions of the Setup error were (0.95 ± 0.89, 1.09 ± 0.94, 1.62 ± 1.18) mm vs (1.02 ± 0.94, 1.14 ± 0.96, 1.82 ± 1.21) mm, p in the z-direction is statistically significant (p = 0.01); The position of the 4,7th cervical vertebra also had statistical significance in the Z direction. Comparison between the surgical study group and non-surgical study group, the errors in the x and Z directions of the clivus and the 7th cervical vertebra were (0.99±0.96, 1.65±1.36 and 1.64±1.23, 1.61±1.37) mm vs (0.80±0.77, 1.85±1.44 and 1.67±1.07, 1.95±1.60) mm, P = 0.02, 0.11, 0.77, 0.01; the results were similar to it when comparing surgical controls to non-surgical controls. It can be seen from the data that in the X and Z directions from the clivus to the neck, the error change of the surgical group is smaller than that of the non-surgical group. There was no difference in toxicity and side effects. <h3>Conclusion</h3> The results show that neck muscle training can improve the anterior and posterior neck deformation error, improve the radiotherapy accuracy and setup efficiency.