2020 Background: Lung cancer is one of the most common causes of brain metastases (BMs) and is always associated with poor prognosis. To evaluate the characteristics of the tumor immune microenvironment in brain metastases of non-small-cell lung cancer (NSCLC), we investigated the immunophenotype of primary NSCLC and paired brain metastases. Methods: Forty-three Chinese patients with NSCLC who had BMs at presentation or during the course of their disease were admitted to the Sun Yat-Sen University Cancer Center (Guangzhou, China) from 2000 to 2019. RNA sequencing (RNA-seq) of eighty-six formalin-fixed, paraffin embedded (FFPE) samples from primary lung tumors and paired brain metastases of 43 patients was conducted to comprehensively analyze the tumor immune microenvironment. Results: Our data revealed that brain metastases compared with primary lung tumors exhibited reduced tumor infiltrating lymphocytes (TILs) (all 28 immune cell subtypes P < 0.05), lower fraction of activated CD8 T cell and effector memory CD8 T cell in total TILs (P = 0.028, P < 0.001, respectively); higher fraction of macrophage and neutrophil in total TILs (P < 0.001, P < 0.01, respectively). Comparing with the primary lung tumors, the scores of some immune related signatures, including MHC non-class signature, IFN gamma signature and T-cell-inflamed gene-expression profile (GEP) signature, were significantly lower in brain metastases (P = 0.004, P = 0.009, P = 0.004, respectively), while the score of MHC class-II signature was higher in brain metastases (P = 0.045). We found the distributions of tumor microenvironment immune types (TMIT) in brain metastases and primary lung tumors were different. Brain metastases contained significantly lower proportion of TMIT I (high PD-L1/ high CD8A) (23%) than primary lung tumors (47%) (P < 0.05). Besides, we found three immune inhibitory checkpoint molecules, namely C10orf54 (VISTA), CTLA4 and CD274 (PD-L1) were downregulated in brain metastases than in primary lung tumors (P < 0.001, P < 0.001, P = 0.034, respectively). Moreover, there was poor correlation of PD-L1 expression between paired brain metastases and primary lung tumors (R = 0.28, P = 0.068). Unsupervised hierarchic cluster analysis revealed the primary lung tumors had two distinct patterns of immune gene signatures, namely Cluster A and Cluster B, and the tumors in Cluster B were immune rich, but associated with poor prognosis (log-rank P = 0.021). Conclusions: Our work illustrates the immune landscape of brain metastases from NSCLC, and suggests that the tumor immune microenvironment in brain metastases compared with primary lung tumors is further immunosuppressed, that may help guide immunotherapeutic strategies for NSCLC brain metastases.