Abstract: This paper presents an approach for extracting complex spatial concepts from unstructured text.By defining several spatial semantic roles based on the characteristics of geo-spatial information in natural language,the authors extract spatial entities,spatial relationships among entities and spatial procedures by means of spatial semantic annotation,semantic phrases recognition and pattern match.It tries to solve the problem of unable to get deep spatial semantics in common IE.The primary experiment shows that it has a good precision and a similar recall comparing to common IE systems.