Some strains of avian influenza A virus (AIV) can directly transmit from their natural hosts to humans. These avian-to-human transmissions have continuously been reported to cause human deaths worldwide since 1997. Predicting whether AIV strains can transmit from avian to human is valuable for early warning of AIV strains with human pandemic potential. In this study, we constructed a computational model to predict avian-to-human transmission of AIV based on physicochemical properties. Initially, ninety signature positions in the inner protein sequences were extracted with the entropy method. These positions were then encoded with 531 physicochemical features. Subsequently, the optimal subset of these physicochemical features was mined with several feature selection methods. Finally, a support vector machine (SVM) model named A2H was established to integrate the selected optimal features. The experimental results of cross-validation and an independent test show that A2H has the capability of predicting transmission of AIV from avian to human.