Kinship verification aims to find out whether there is a kin relation for agiven pair of facial images. Kinship verification databases are born withunbalanced data. For a database with N positive kinship pairs, we naturallyobtain N(N-1) negative pairs. How to fully utilize the limited positive pairsand mine discriminative information from sufficient negative samples forkinship verification remains an open issue. To address this problem, we proposea Discriminative Sample Meta-Mining (DSMM) approach in this paper. Unlikeexisting methods that usually construct a balanced dataset with fixed negativepairs, we propose to utilize all possible pairs and automatically learndiscriminative information from data. Specifically, we sample an unbalancedtrain batch and a balanced meta-train batch for each iteration. Then we learn ameta-miner with the meta-gradient on the balanced meta-train batch. In the end,the samples in the unbalanced train batch are re-weighted by the learnedmeta-miner to optimize the kinship models. Experimental results on the widelyused KinFaceW-I, KinFaceW-II, TSKinFace, and Cornell Kinship datasetsdemonstrate the effectiveness of the proposed approach.