The invention discloses a user portraying method based on mass cross-screen behavior data. The method mainly comprises the steps that in the face of bidirectional new multimedia business which is increasingly flexible and millions and even ten millions of mass behavior data of users, collected user behavior data is subjected to HDFS distributed storage; after the data is extracted, converted and loaded through an ETL module, an optimal combinational algorithm conforming to the characteristics of the media industry is used to perform efficient data preprocessing on mass user behavior data fused with content tags, user tags, consumption tags, geotags, equipment tags, user properties etc.; and finally, user portraits are formed, and then relevant user portraits are called through a WEB application. The method provides precise data support for business operation by a broadcast television network operator.