Glioblastoma multiforme (GBM) is a common lethal brain tumor. Genes involved in the pathogenesis of GBMs may have diagnostic value or be used as important prognostic factors. We have profiled whole genome mRNA expression array in 119 GBM patients and 5 normal brain tissues and transcriptome sequencing in 98 GBM patients from the Chinese Glioma Genome Atlas (CGGA) cohort to identify prognostic markers for GBM patients. Multivariate Cox proportional hazard regression analysis (p<0.01) and Significance Analysis of Microarrays (SAM, FDR<5%) were performed to identify prognostic genes in the GBMs group. Meanwhile, gene expression profile data of 196 GBMs from The Cancer Genome Atlas (TCGA) network was utilized to validate our results. Only genes significant in both CGGA and TCGA databases were chosen for further analysis. Kaplan-Meier survival analysis was performed to estimate overall survival of all GBM patients. PAX3 was identified as the most significantly prognostic gene using the above methods. Overexpression of PAX3 indicated poor survival in the 183 CGGA IDH1 wild type GBM patients but not in those of TCGA IDH1 wild type GBM patients. Prognostic value of PAX3 overexpression in GBMs was not interfered by clinicopathological features. Functional annotation of mRNA profiling and transcriptome data in GBMs from the CGGA and TCGA cohorts showed that patients with higher PAX3 expression tended to have a higher expression of proliferation associated genes. Knocking down the expression of PAX3 in U87 glioblastoma cells with PAX3 siRNA resulted in significant decrease in GBM cell proliferation and migration. The present study suggests that overexpression of PAX3 could be a marker with poor prognosis in IDH1 wild type GBMs by increasing cancer cell proliferation and migration.