In clinical practice, the detection of biomarkers is mostly based on primary tumors for its convenience in acquisition. However, immune checkpoints may express differently between primary and metastatic tumor. Therefore, we aimed to compare the differential expressions of PD-1, PD-L1 and PD-L2 between the primary and metastatic sites of renal cell carcinoma (RCC). Patients diagnosed with RCC by resection or fine needle aspiration of metastasis were included. Immunohistochemistry (IHC) was applied to detect PD-1, PD-L1 and PD-L2 expressions. SPSS 22.0 was applied to conduct Chi-square, consistency tests and Cox's proportional hazards regression models. GraphPad Prism 6 was used to plot survival curves and R software was used to calculate Predictive accuracy (PA). In the whole cohort (N = 163), IHC results suggested a higher detection rate of PD-L1 in the metastasis than that of the primary site (χ2 = 4.66, p = 0.03), with a low consistent rate of 32.5%. Among different metastatic tumors, PD-1 was highly expressed in the lung/lymph node (65.3%) and poorly expressed in the brain (10.5%) and visceral metastases (12.5%). PD-L1 was highly expressed in lung/lymph node (37.5%) and the bone metastases (12.2%) on the contrary. In terms of survival analysis, patients with PD-1 expression either in the primary or metastasis had a shorter overall survival (OS) (HR: 1.59, 95% CI 1.08-2.36, p = 0.02). Also, PD-L1 expression in the primary was associated with a shorter OS (HR 2.55, 95% CI 1.06-6.15, p = 0.04). In the multivariate analysis, the predictive accuracy of the whole model for PFS was increased from 0.683 to 0.699 after adding PD-1. PD-1, PD-L1 and PD-L2 were differentially expressed between primary and metastatic tumors. Histopathological examination of these immune check points in metastatic lesions of mRCC should be noticed, and its accurate diagnosis may be one of the effective ways to realize the individualized treatment.