Background: In this study, we evaluated the difference in the tumor mutational burden (TMB) score between primary and metastatic lesions in pan-cancer and the different cut-offs for guiding immune checkpoint inhibitors (ICIs) prognosis. Methods: We screened 1661 pan-cancer cases from the cBioPortal database. The Kaplan-Meier method was applied to obtain survival curves that were compared using the log-rank test. The X-tile model was used to determine the optimal cut-off values of TMB. Results: Our results showed that the tissues obtained from the metastatic lesions had more co-occurring gene mutations than the primary lesions (P<0.05, Q<0.05). Moreover, tissues from patients with metastatic lesions had a higher TMB score than those from patients with primary lesions (P = 0.024). According to the median cut-off values of TMB from the primary and metastatic lesions, we analyzed the overall survival (OS) in the low TMB and high TMB groups, respectively. The results showed different OS for the two TMB groups in primary and metastatic lesions. Subsequently, we analyzed the optimal cut-off values of TMB score to predict survival in primary and metastatic lesions based on the X-tile model. The optimal cut-off value for the test tissue from primary lesions was 20.19 (P<0.001). Importantly, the optimal cut-off value for test tissue from metastatic lesions was 10.18 (P < 0.001). Conclusion: TMB detection from primary and metastatic lesions should be considered separately to predict survival for ICIs treatment.