A new approach to improve the reliability of dynamic susceptibility contrast MRI for the evaluation of brain tumor hemodynamics in the presence of contrast agent extravasation is described. This model-based technique simultaneously estimates the voxel-wise tumor residue function and the temporal extravascular T(1) changes following contrast agent leakage. With these estimates the model corrects the measured MRI signal, which is then used to calculate tumor hemodynamic parameters. The feasibility of this technique is demonstrated with computer simulations that cover a wide range of hemodynamic conditions and by application to eight tumor-bearing rats. The simulations demonstrate that the corrected hemodynamic parameters precisely matched the actual values with a maximum percentage error of 4.2% compared to 68.6% for the uncorrected parameters. The corrected parameters are also essentially independent of the tumor hemodynamic state and degree of contrast extravasation. Consistent with these improvements, significant differences between corrected and uncorrected parameters, calculated from a gradient-echo sequence, are shown in a rat 9L gliosarcoma model. This method combined with the hemodynamic parameters derived from GE and SE sequences shows promise as a new tool to evaluate tumor angiogenesis and its therapy.