Background: Glioblastoma (GBM) is the most common and lethal primary brain tumor in adults. Several recent transcriptomic studies in GBM have identified different signatures involving immune genes associated with GBM pathology, overall survival (OS) or response to treatment. Methods: In order to clarify the immune signatures found in GBM, we performed a co-expression network analysis that grouped 791 immune-associated genes (IA genes) in large clusters using a combined dataset of 161 GBM specimens from published databases. We next studied IA genes associated with patient survival using 3 different statistical methods. We then developed a 6-IA gene risk predictor which stratified patients into two groups with statistically significantly different survivals. We validated this risk predictor on two other Affymetrix data series, on a local Agilent data series, and using RT-Q-PCR on a local series of GBM patients treated by standard chemo-radiation therapy. Results: The co-expression network analysis of the immune genes disclosed 6 powerful modules identifying innate immune system and natural killer cells, myeloid cells and cytokine signatures. Two of these modules were significantly enriched in genes associated with OS. We also found 108 IA genes linked to the immune system significantly associated with OS in GBM patients. The 6-IA gene risk predictor successfully distinguished two groups of GBM patients with significantly different survival (OS low risk: 22.3 months versus high risk: 7.3 months; p < 0.001). Patients with significantly different OS could even be identified among those with known good prognosis (methylated MGMT promoter-bearing tumor) using Agilent (OS 25 versus 8.1 months; p < 0.01) and RT-PCR (OS 21.8 versus 13.9 months; p < 0.05) technologies. Interestingly, the 6-IA gene risk could also distinguish proneural GBM subtypes. Conclusions: This study demonstrates the immune signatures found in previous GBM genomic analyses and suggests the involvement of immune cells in GBM biology. The robust 6-IA gene risk predictor should be helpful in establishing prognosis in GBM patients, in particular in those with a proneural GBM subtype, and even in the well-known good prognosis group of patients with methylated MGMT promoter-bearing tumors.