Aiming at the problem that inevitable outliers occur in GNSS position and velocity for land vehicular navigation under complex urban environment, which would deteriorate the estimation accuracy of GNSS/MEMSINS navigation state parameters and even lead to the filtering divergence, in this paper an adaptively outlierrestrained GNSS/MEMSINS integrated navigation algorithm is proposed to improve the accuracy and reliability of integrated navigation based on the faulttolerant adaptive Kalman filtering. This algorithm establishes accurate noise model for MEMSbased inertial sensors with Allan variance analysis technique, which reduces the influence of kinematic model mismatch and state disturbances effectively. The innovation sequences are used to construct the test statistic for detecting observation outliers. The adaptive innovation weighting factor is constructed according to the statistic to adjust the filter gain matrix, and weaken the adverse influence of observation outliers on state estimation. The field test result indicates that the proposed algorithm can effectively control the influences of GNSS position and velocity outliers, and has strong realtime and faulttolerant ability for GNSS/MEMSINS integration navigation. The estimation accuracies of position, velocity and attitude determination are improved by 35.78%, 60.19% and 82.41%, respectively compared with those of traditional algorithm, which verifies the effectiveness and practicability of the proposed algorithm.