Rotation estimation of known rigid objects is important for robotic applications such as dexterous manipulation. Most existing methods for rotation estimation use intermediate representations such as templates, global or local feature descriptors, or object coordinates, which require multiple steps in order to infer the object pose. We propose to directly regress a pose vector from point cloud segments using a convolutional neural network. Experimental results show that our method achieves competitive performance compared to a state-of-the-art method, while also showing more robustness against occlusion. Our method does not require any post processing such as refinement with the iterative closest point algorithm.