Brain metastasis is an extremely serious sequela with a dismal prognosis in non-small cell lung cancer (NSCLC). The present study aimed to identify novel biomarkers and potential therapeutic targets for brain metastases of NSCLC. We performed high-throughput Luminex assays to profile the transcriptional levels of 36 genes in 70 operable NSCLC patients, among whom 37 developed brain metastases as the first relapse within 3 years after surgery. The Cox proportional hazards regression model was used to evaluate the association between genes and brain metastases. Wound healing assay and transwell assay was carried out to estimate the function of target gene in vitro. And left ventricular injection on nude mice was used to evaluate the effect of target gene in vivo. Growth-associated protein 43 (GAP43) was found to be related to brain metastasis. Multivariate Cox regression analysis showed that NSCLC patients with elevated GAP43 had a 3.29-fold increase in the risk for brain metastasis compared with those with low levels (95% confidence interval: 1.55–7.00; P = 0.002). Kaplan–Meier survival curves revealed that GAP43 was also associated with overall survival. Analysis of a cohort of 1926 NSCLC patients showed similar results: patients with high levels of GAP43 had worse progression-free and overall survival rates. Furthermore, in vitro experiments showed that GAP43 facilitated cell migration. Animal studies demonstrated that GAP43-silenced NSCLC cells were less likely to metastasize to the brain and bone than control cells. Immunofluorescence and F-actin/G-actin in vivo assays indicated that GAP43 knockdown triggered depolymerization of the F-actin cytoskeleton. Rho GTPase activation assays showed that Rac1 was deactivated after GAP43 was silenced. Our findings suggest that GAP43 is an independent predictor of NSCLC brain metastasis and that it may facilitate metastasis by regulating the Rac1/F-actin pathway.