Children with brain arteriovenous malformations (bAVM) are at risk of life-threatening hemorrhage contributing to unfavorable neurological deficit in their early lives. Our aim was to propose a classification system predicting the unfavorable hemorrhage in children with bAVM.We identified all consecutive children admitted to our institution for bAVMs between July 2009 and August 2015. A hemorrhage event was defined as unfavorable when it is life-threatening (requiring emergent invasive intervention) or with post-hemorrhage mRS > 3. The effects of demographic characteristics and bAVM morphology on unfavorable hemorrhage risk were studied using univariate and multivariable regression analyses, followed by discrimination analysis using area under the receiver operating curve (AUROC) and 5-fold cross validation.A total of 162 pediatric bAVM cases were identified, unfavorable hemorrhage occurred in 49 (30.2%). Periventricular nidus location (HR, 4.46; 95%CI, 1.93-10.31; P < 0.001), non-temporal lobe location (HR, 2.72; 95%CI, 1.20-6.15; P = 0.02) and long pial draining vein (HR, 3.26; 95%CI, 1.53-6.97; P = 0.002) were independent predictors of an earlier unfavorable hemorrhage in pediatric bAVMs. We further classified the bAVM into three types: Type I, periventricular and non-temporal location (Ia, deep location; Ib, superificial location); Type II, with long pial draining vein and non-periventricular or temporal location; Type III, non-periventricular or temporal location without long draining vein. Predictive accuracy of this classification for unfavorable hemorrhage was assessed with AUROC of 0.77 (95% CI 0.69-0.85) and remained stable after cross validation.A morphologic model based on nidus location and venous drainage might predict unfavorable hemorrhage in children with bAVM.