Attributed network embedding aims to learn low-dimensional node representations from both network structure and node attributes. Existing methods can be categorized into two groups: (1) the first group learns two separated node representations from network structure and node attribute respectively and concatenating them together; (2) the other group obtains node representations by translating node attributes into network structure or vice versa. However, both groups have their drawbacks. The first group neglects the correlation between these two types of information, while the second group assumes strong dependence between network structure and node attributes. In this paper, we address attributed network embedding from a novel perspective, i.e., learning representation of a target node via modeling its attributed local subgraph. To achieve this goal, we propose a novel graph auto-encoder framework, namely GraphAE. For a target node, GraphAE first aggregates the attribute information from its attributed local subgrah, obtaining its low-dimensional representation. Next, GraphAE diffuses its representation to nodes in its local subgraph to reconstruct their attribute information. Our proposed perspective transfroms the problem of learning node representations into the problem of modeling the context information manifested in both network structure and node attributes, thus having high capacity to learn good node representations for attributed network. Extensive experimental results on real-world datasets demonstrate that the proposed framework outperforms the state-of-the-art network approaches at the tasks of link prediction and node classification.