The capacity of attention is crucial for robotic manipulator performing many tasks. This paper presents an Emotion-Driven Attention(EDA) model for robotic manipulator action selection, which integrates emotion with the robotic manipulator to manage its attention allocation, and enable it react to emotional social cue appropriately. During the interaction with human, the human's facial expression can be recognized by the robot in the real-time, which is considered as a social cue to trigger a corresponding emotion of the robot by Self-Organizing Map (SOM). Then, the robot's emotion plays as a reinforcement signal to regulate the attention parameters, guiding the robot's visual attention on the identified object and making it behave appropriately according to the human intentions, e.g., grasping or avoiding the object. We estimate the proposal in both simulated situations and interactive scenarios, the robot's simulated attention intensity and the attention behaviors driven by different emotions indicates the effectiveness of the proposal.