OBJECTIVE Case selection for the surgical treatment of brain arteriovenous malformations (BAVMs) remains challenging. This study aimed to construct a predictive grading system combining lesion-to-eloquence distance (LED) for selecting patients with BAVMs for surgery. METHODS Between September 2012 and September 2015, the authors retrospectively studied 201 consecutive patients with BAVMs. All patients had undergone preoperative functional MRI and diffusion tensor imaging (DTI), followed by resection. Both angioarchitectural factors and LED were analyzed with respect to the change between preoperative and final postoperative modified Rankin Scale (mRS) scores. LED refers to the distance between the lesion and the nearest eloquent area (eloquent cortex or eloquent fiber tracts) measured on preoperative fMRI and DTI. Based on logistic regression analysis, the authors constructed 3 new grading systems. The HDVL grading system includes the independent predictors of mRS change (hemorrhagic presentation, diffuseness, deep venous drainage, and LED). Full Score combines the variables in the Spetzler-Martin (S-M) grading system (nidus size, eloquence of adjacent brain, and venous drainage) and the HDVL. For the third grading system, the fS-M grading system, the authors added information regarding eloquent fiber tracts to the S-M grading system. The area under the receiver operating characteristic (ROC) curves was compared with those of the S-M grading system and the supplementary S-M grading system of Lawton et al. RESULTS LED was significantly correlated with a change in mRS score (p < 0.001). An LED of 4.95 mm was the cutoff point for the worsened mRS score. Hemorrhagic presentation, diffuseness, deep venous drainage, and LED were independent predictors of a change in mRS score. Predictive accuracy was highest for the HDVL grading system (area under the ROC curve 0.82), followed by the Full Score grading system (0.80), the fS-M grading system (0.79), the supplementary S-M grading system (0.76), and least for the S-M grading system (0.71). Predictive accuracy of the HDVL grading system was significantly better than that of the Spetzler-Martin grade (p = 0.040). CONCLUSIONS LED was a significant predictor for the preoperative risk evaluation for surgery. The HDVL system was a good predictor of neurological outcomes after BAVM surgery. Adding the consideration of the involvement of eloquent fiber tracts to preoperative evaluation can effectively improve its predictive accuracy.