In　finger-vein-based　biometric　verification,　it　is　essential　to　robustly　extract　vein　features　with　strong　discrimination　ability.　Recently,　deep　learning　methods　have　achieved　remarkable　performance　in　the　field　of　finger-vein　verification.　However,　the　establishment　of　an　effective　deep　learning　model　requires　large-scale　databases　to　prevent　overfitting　during　the　training　process,　while　currently　used　finger-vein　databases　are　not　large　enough.　In　our　article,　a　new　generative　adversarial　network　(GAN),　named　triplet-classifier　GAN,　is　designed　for　finger-vein　verification.　Unlike　the　traditional　GAN-based　method,　the　proposed　triplet-classifier　GAN　uses　the　generated　＂fake＂　data　to　improve　the　learning　ability　of　the　triplet　loss-based　convolutional　neural　network　(CNN)　classifier.　The　combination　of　GAN　and　the　triplet　loss-based　CNN　classifier　expands　the　training　data　and　improves　the　discriminant　ability　of　CNN.　Experiments　prove　that　the　proposed　triplet-classifier　GAN　has　superior　performance　in　finger-vein　verification　and　has　good　prospects　in　finger-vein-based　biometric　verification.