In this paper, we present a fast and fully automatic learning based system that is capable of detecting coronary stenoses in Computed Tomography Angiogra-phy (CTA) caused by all types of plaques, e.g. soft, mixed, and calcified. We extract geometrical and intensity based features that can capture the characteristic properties of the coronary vessels. We evaluated our method on the Rotterdam Coronary Artery Stenoses Detection and Quantification Evaluation Framework on 42 datasets. On the 24 testing datasets, a sensitivity of 57% and a PPV of 18% is achieved as compared to QCA, while a sensitivity of 57% and a PPV of 32% is achieved as compared to CTA. This clearly indicates that our method is good at ruling out disease (low false
negative detection value), but has limited performance to detect significant stenoses (> 50% luminal diameter reduction; high false positive rate).