Information Extraction from semistructured data becomes more and more important. In order to extract meaningful or interesting contents from semistructured data, we need to extract common structured patterns from semistructured data. Many semistructured data have irregularities such as missing or erroneous data. A tag tree pattern is an edge labeled tree with ordered children which has tree structures of tags and structured variables. An edge label is a tag, a keyword or a wildcard, and a variable can be substituted by an arbitrary tree. Especially, a contractible variable matches any subtree including a singleton vertex. So a tag tree pattern is suited for representing common tree structured patterns in irregular semistructured data. We present a new method for extracting characteristic tag tree patterns from irregular semistruc-tured data by using an algorithm for finding a least generalized tag tree pattern explaining given data. We report some experiments of applying this method to extracting characteristic tag tree patterns from irregular semistructured data.