Arterial Spin Labeling (ASL) is a recent MRI perfusion technique which enables quantification of cerebral blood flow (CBF). The presence of regions with atypical CBF can characterize a pathology. In brain tumors for instance, perfusion increase can be directly related to the grading of the malignant tissues. It is therefore of great interest to identify these regions in order to provide the patients with the most appropriate therapy. In this paper, we propose to detect abnormal brain perfusion by using an a contrario framework and an ASL template as a model of normal perfusion. Validation was undertaken by qualitative comparison with CBF extracted from dynamic susceptibility weighted contrast enhanced (DSC) imaging. We experimented this framework on four patients presenting brain tumors. Results show that high perfusion regions found in DSC CBF maps are correctly identified as hyperperfusions with a contrario detection. Automatic detection has clear advantages over manual delineation since it is less time-consuming, does not depend on medical expertise and allows quantification of perfusion abnormalities within the detected regions