Face alignment is a key component of face recognition system, and facial landmark points are widely used for face alignment by a number of face recognition systems. However, inaccurate locations of landmark points bring about spatial misalignment which degrades the performance of face recognition systems. In order to alleviate this problem, we propose a simple and efficient data augmentation approach, which uses artificial landmark perturbation to generate a huge number of misaligned face images, to train Deep Convolutional Neural Networks (DCNN) models robust to landmark misalignment. In our experiments, three types of facial landmark-based face alignment methods are applied to train DCNN models on CASIA-WebFace training database. Experimental results on Labeled Faces in the wild database (LFW) and YouTube Faces database (YTF) verify the effectiveness of our approach. (C) 2016 Elsevier B.V. All rights reserved.