In this paper, we describe our approach of automatically extracting property-value pairs from the Web for organizations when only the name and address information are known. In order to explore the enormous knowledge from the Web, we first retrieve the Web pages containing organization properties by search engine, and then automatically extract the property-value pairs regardless of heterogeneous Web page structures. Our method does not require any training data or human-made template. We have constructed an organization knowledge base containing 3 million entities extracted from the Web for 4.2 million organizations which only have name and address information. The experiment shows that our approach makes it possible and effective for people to construct their own knowledge base.