For target tracking in wireless sensor networks under cyber attacks, this paper presents a Gaussian mixture distributed robust set-membership filtering algorithm, aiming to improve the consistency and accuracy of moving target tracking under malicious cyber attacks. The algorithm can be decomposed into three steps: the correction/measurement update step, the clustering fusion step, and the prediction/time update step. The correction/measurement update step is used to update the predicted state estimation (a priori estimation) based on the local measurement. In the clustering fusion step, the available local estimations of sensor nodes are clustered by the Gaussian mixture model clustering algorithm, which are classified into trusted nodes estimations and non-trusted nodes estimations. The non-trusted nodes estimations are ignored while the trusted nodes estimations are used to generate the fused estimation. The prediction/time update step is used for predicting the priori state estimate of the moving target and pass the current moment state estimate of the target to the next moment. Simulation results show that the proposed algorithm is robust against five common cyber attacks, namely random attacks, denial of service attacks, false data injection attacks, replay attacks and hybrid attacks.