Biometric recognition, which is an integral part of the present-day security solutions, faces a major threat from presentation or spoofing attacks. In this paper, we present a novel presentation attack detection (PAD) approach for dorsal hand-vein based recognition system. The proposed approach performs Laplacian of Gaussian filtering on the acquired images, followed by extraction of histogram of oriented gradients (HoG) features at multiple scales. A linear SVM is employed for each scale and the final decision is obtained by combining individual decisions using the majority voting scheme. Experiments were carried out on 624 real images and 624 artefacts (spoof samples) collected from left and right hands of 52 subjects. Artefacts were generated independent of the enrollment images, by employing an off-the-shelf smartphone camera to capture the vein patterns from users' hands. These images were displayed using two different display devices and presented as artefacts to the biometric sensor. The experiments were carried out in the same-device and the cross-device scenarios. Our approach achieves average error rate of 0.16% and 0.8% in the same-device and the cross-device experiments, respectively and outperforms local binary patterns (LBP) based baseline algorithm.