Background: Anti-angiogenic therapies, anti-VEGF agents, cause high radiographic response rates due to reduction in vascular permeability. However, we theorized the degree of tumor shrinkage combined with the duration of response might be a more meaningful measure of tumor control and may better reflect real patient benefit. In the current study, we investigate the association between model-derived parameters describing enhancing tumor volumetric dynamics and overall survival (OS) in recurrent glioblastoma treated with anti-VEGF therapy from multiple clinical trials. Methods: N=134 patients treated with bevacizumab +/- irinotecan (BRAIN Trial; NCT00345163) and N=121 patients treated with cabozantinib (XL184-201; NCT00704288) were used to identify potential imaging biomarkers, and N=63 patients in the bevacizumab control arm in a phase III trial (GLOBE Trial; NCT02511405) were used as a validation dataset. Enhancing tumor volumes were estimated using T1 subtraction maps, and a biexponential model was used to model tumor response (d) and regrowth (g) rates, as well as time to tumor regrowth (TTG), the inflection point where the tumor stops shrinking and starts to regrow, and the depth of response (DpR) after initiation of anti-VEGF therapy for each patient. Log-rank, univariate, and multivariable Cox regression were used to quantify the relationship of these parameters and OS. Results: Optimized thresholds based on maximum hazard ratio occurred at d=0.11 months^-1 (P=5x10^-9 training; P=0.0082 validation); g=0.07 months^-1 (P=5x10^-20 training; P<0.0001 validation); TTG of 3.8 months (P=6x10^-17training; P<0.0001 validation); and DpR of 11.3% (P=0.0026 training; P=0.0177 validation). Univariate and multivariable Cox regression using these thresholds and controlling for age and baseline tumor volume confirmed that d, g, TTG, and DpR were all significant and independent predictors of OS in both phase II training and phase III validation datasets (P<0.01). Additionally, patients exhibiting combinations of a high DpR and/or long TTG showed differential median OS, while patients with favorable diffusion MR phenotypes exhibited a longer OS and significantly longer TTG. Conclusions: Volumetric response and regrowth rates, TTG, and DpR using a biexponential model are significant and independent predictors of OS in recurrent glioblastoma treated with anti-VEGF therapy. Clinical trial information: NCT00345163; NCT00704288; NCT02511405.