Medulloblastoma (MB) is the most common brain tumor in children. There are four distinct subtypes of MB, but patients with anaplastic/large cell have the worst prognosis. Since the morbidity is highly correlated with treatment for MB, the ability to distinguish aggressive (such as anaplastic/large cell) MB is crucial. We present a scheme that leverages quantitative image texture features (Haar, Haralick, and Laws) and classifier ensembles (random forests) to automatically classify histological images from MB resection as being anaplastic/large cell or non-anaplastic/large cell. Preliminary results for our scheme when applied to patch-based classification of MB specimens yield an AUC of 0.91.