Abstract: A data model, named data tree, is proposed as the common data model for integration of multiple sources, especially the sources without predictable data schemata. This class of sources includes databases with often changing schemata, data sources with unstructured or semistructured data, and data sources with self describing data. In contrast to most data models, metadata of data tree are not stored separately but coexist with data in a tree structure. As the mathematical foundation of the manipulation language of data tree, a new algebra, called data tree algebra, which includes eight operations, is developed. In comparison with relational algebra, data tree algebra is more flexible and powerful.