Semi-structured data are typically represented in the form of labeled directed graphs. They are self-describing and schemaless. The lack of a schema renders query processing over semi-structured data expensive. To overcome this predicament, some researchers proposed to use the structure of the data for schema representation. Such schemas are commonly referred to as graph schemas. Nevertheless, since semi- structured data are irregular and frequently subjected to modifications, it is costly to construct an accurate graph schema and worse still, it is difficult to maintain it thereafter. Furthermore, an accurate graph schema is generally very large, hence impractical. In this paper, an approximation approach is proposed for graph schema extraction. Approximation is achieved by summarizing the semi-structured data graph using an incremental clustering method. The preliminary experimental results have shown that approximate graph schemas were more compact than the conventional accurate graph schemas and promising in query evaluation that involved regular path expressions.