Abstract: The flexible nomenclature of gene name results in severe semantic ambiguity, which is an obstacle for deep biomedical text mining. Gene name normalization (GN) is an effective way to resolve this problem. In this work, a multi-level disambiguation framework was proposed to solve gene name normalization problem. Aiming at different ambiguity situations during the procedure of GN, three different strategies were included in the framework. They were dictionary-based gene name detection, machine-learning-based candidate selection, and semantic-based disambiguation. Experimental results showed that the proposed method could achieve 0.746 F-measure on the BioCreAtIvE2006 GN task test data set.