Avian influenza A virus (AIV) can cross the host barrier to infect human directly and has continuously been reported to cause human death worldwide. Predicting which AIVs can directly transmit from avian to human will provide early warning of AIVs with human pandemic potential, which is beneficial to public health. Although it is easy to decide a dataset of AIVs having the capability of avian-to-human transmission as positive samples, there are no experimentally confirmed AIVs without the capability to be considered as negative samples. Therefore, in this study we utilized one-class support vector machines (OCSVM) to solve this one-class classification problem. With two feature sets including amino acid composition and Moran autocorrelation, an OCSVM-based prediction model was constructed and demonstrated to achieve good performances on both the training dataset and the external testing dataset. The experimental results imply that the model constructed on only positive samples (AIVs having the capability of avian-to-human transmission) is efficient to predict avian-to-human transmission of AIVs.