Effective document classification is a long-pursued goal in knowledge management. This paper proposes a novel hybrid approach of semantic representation and statistical measurements. Document is divided into content segments first. By Formal Concept Analysis (FCA), their semantic links with standard concept identifiers are built up whose weights are calculated statistically. In this way, effective concept fusing and document classification can be achieved. In addition, a semantic overlay for specific documents will be constructed via concept fusing. Experiments show our approach is feasible and effective.