Pathogenic avian and human influenza virus often cause disastrous damage to human society and economics. Understanding antigenic drift of influenza viruses is an emergent issue for prophylaxis and vaccine development. In this study, we identified antigenic critical amino acid positions on hemagglutinin (HA) gene and rules for predicting antigenic variants. The information gain (IG) is applied to calculate the degree of association between the position mutation and antigenic drift. An amino acid with high IG at a specific position implied that this position is highly correlated to antigenic type change. The decision tree is applied to build model and discover rules for predicting antigenic variants. The decision tree can identify amino acid mutation rules that lead to the antigenic drift. The predicting accuracy of this model is 91.2% which is better than related models. Most of these selected positions with high IG are located on epitope or surface on the HA structure. The position 145 with highest information gain could lead to antigenic cluster transition were verified. These results demonstrate that our approach is robust and is potential useful for vaccine development.