Target-dependent sentiment classification aims at identifying the sentiment polarities of targets in a given sentence. Previous approaches utilize recurrent neural network with attention mechanism incorporated to model the context and learn key sentiment intermediate representation in relation to a given target. However, such methods are incapable either of modeling complex contexts or of processing data parallelly. To address these problems, we propose, in this paper, a new model that employs a multi-layer convolutional neural network to process the context parallelly and model the context multiple times, where the neural network is able to explicitly learn the sentiment intermediate representation via an attention mechanism. Eventually, we integrate these features to form a final sentiment representation, which will be fed into the classifier. Experiments show that our model surpasses the existing approaches on several datasets.