The invention, which belongs to the field of communication network security and machine learning, discloses a method for analyzing a fraudulent phone number. The method includes: analyzing a behavioral characteristic and a behavior scene in a phone fraud event according to relevant fraud case experience and original call bill information of various telecommunication fraud events, and carrying out screening on original call bills for different behavior scenes; extracting and establishing multi-dimensional feature sets for different fraud types; establishing different types of fraud phone detection models by using a random forest algorithm, carrying out offline training on fraud phone number classification models periodically; carrying out quasi real-time detection and determination in an actual network; identifying a fraud phone event and a fraud phone number form lots of call bill information; and then carrying out reporting and processing on the fraud phone event and the fraud phone number by taking a corresponding way to maintain security of the telecommunications network. Therefore, a fraud phone number in communication can be identified effectively and accurately.