The invention relates to an information maintenance technology in the field of computers, in particular to a method for cleaning junk data. The method comprises the steps of establishing a junk data cleaning rule system and detecting error data and repeating data from a data source so as to filter information of a database. The method has the following benefits: 1, the rules have expansibility, so a new rule can be supplemented at any time; 2, the rules and the data form a many-to-many mapping relationship so as to allow one data item to correspond to multiple rules and also allow one rule to correspond to multiple data items; 3, the rules are semantic, so the established rules are shown in an easily understood form and stored in a computer language manner; 4, the junk data processing scheme is configured flexibly and allows to process the junk data in multiple self-defining modes comprising deleting, jumping, manual processing and the like; and 5, multi-database support is adopted so as to allow to process support of main stream databases comprising oracle, mysql, db2 and the like.