The solution to semantic heterogeneity of the integrated data is a key point in the integration of heterogeneous databases. To aim at adding semantic information to tables and fields in relational databases, a semantic annotation method was developed. The method integrates many different string similarity algorithms to improve similarity algorithms. Firstly, extracting implicit semantic information from metadata in relational databases according to similarities between metadata and the ontology entity; then, the extracted information is annotated using name similarities and structural similarities between metadata and the ontology entity. Tests annotating the metadata in dozens of heterogeneous relational databases had an accurate rate of 82%.