Background: Despite decades of clinical trials, glioblastoma (GBM) remains a rapidly fatal disease. Much of therapy resistance in GBM is attributed to intratumoral heterogeneity in which subclones rapidly evolve and treatment-resistant cell populations emerge. Understanding heterogeneity at diagnosis and its relationship to the origins of recurrence are critical to selecting therapies that are efficacious across the entire tumor. However, few genomic studies go beyond analyzing single tumor samples per patient. Methods: A spatially oriented, whole tumor sampling approach was used to obtain 43 biopsies from 3 GBMs at diagnosis and recurrence. Pyclone and ClonEvol were applied to whole exome sequencing to reconstruct clonal evolution, FACETs identified copy number variations and HATCHet estimated tumor purity. Candidate driver mutations were evaluated in relation to the Catalogue of Somatic Mutations In Cancer. Results: A single founding clone and multiple subclones were identified for each diagnosis-recurrence pair. Tumor-wide clonal alterations representing initial clonal expansions of these GBMs included both canonical changes (Chr 7 gain, Chr 10 loss, CDKN2A deletion, EGFR amplification) and a diverse set of large-scale copy number variations (Chr 19, 20 gain), driver mutations (PTEN, KDR, CDH11, CNTNAP2), and fusions (LIMCH1::UCHL1, KANK::DOCK8). A second subset of alterations (Chr 8 gain, ATRX mutation), appeared to be tumor wide at diagnosis but were not identified at recurrence. Cancer drivers were also present subclonally, including CDKN2A deletion, MDM2 amplification, and mutations in NF1, and GRM3. Evolutionary trees consisted of 5 generations of clones in Patient 323 (P323), 3 in P454, and 4 in P534. Divergence of the recurrent tumors from their matched primary occurred in the second generation in P454 and P534 and in the third in P323. As a result, an average of 37% of potential drivers of oncogenesis and clonal expansion across the cohort appear after divergence. Furthermore, each recurrent tumor contained at least one tumor wide driver alteration that was subclonal or undetected at diagnosis. Conclusions: Whole-tumor sampling of three GBM patients at both diagnosis and recurrence identified a diversity of genomic drivers and deeper and more complex genetic roots of individual GBM than previously seen in single-biopsy studies. Tumor recurrences consistently arose from a single subclone that diverged early in evolution of the primary tumor and contained clonal drivers either not detected or subclonal in the primary—suggesting a role for these drivers in persistence and expansion. In the age of personalized medicine, our study highlights the clinical potential of tumor wide sampling in identifying therapeutic targets that could avoid heterogeneity-related therapeutic failures.