Topic maps are a Semantic Web technology that provides a human-oriented mechanism to encode knowledge by organizing web information around topics. Studies have shown, however, that authors face major difficulties in constructing topic maps. This paper discusses an approach to automatic construction of a “draft” topic map for the authors to start with. The idea is to extract topic map constructs by crawling a website and parsing its pages. We propose a set of heuristics that can be used for extracting semantic information from the HTML markup of the web pages. We have used this approach to design and implement a plug-in for the topic map editor TM4L that automatically extracts topics and relationships from a website specified by the author. An evaluation of the proposed approach in terms of Recall and Precision of the extracted data is presented.