Cell line models have been widely used to investigate glioblastoma multiforme (GBM) pathobiology and in the development of targeted therapies. However, GBM tumours are molecularly heterogeneous and how cell lines can best model that diversity is unknown. In this report, we investigated gene expression profiles of three preclinical growth models of glioma cell lines, in vitro and in vivo as subcutaneous and intracerebral xenografts to examine which cell line model most resembles the clinical samples. Whole genome DNA microarrays were used to profile gene expression in a collection of 25 high-grade glioblastomas, and comparisons were made to profiles of cell lines under three different growth models. Hierarchical clustering revealed three molecular subtypes of the glioblastoma patient samples. Supervised learning algorithm, trained on glioma subtypes predicted the intracerebral cell line model with one glioma subtype (r = 0.68; 95% bootstrap CI –0.41, 0.46). Survival analysis of enriched gene sets (P < 0.05) revealed 19 biological categories (146 genes) belonging to neuronal, signal transduction, apoptosis- and glutamate-mediated neurotransmitter activation signals that are associated with poor prognosis in this glioma subclass. We validated the expression profiles of these gene categories in an independent cohort of patients from ‘The Cancer Genome Atlas’ project (r = 0.62, 95% bootstrap CI: –0.42, 0.43). We then used these data to select and inhibit a novel target (glutamate receptor) and showed that LY341595, a glutamate receptor specific antagonist, could prolong survival in intracerebral tumour-implanted mice in combination with irradiation, providing an in vivo cell line system of preclinical studies.