The problem of robust mainlobe deceptive jammer suppression in an element-pulse coding-multiple-input–multiple-output radar is explored utilizing data-dependent beamforming. In general, by introducing a tunable coding coefficient in both transmit elements and pulses, the response of mainlobe deceptive jammers can be identified as signals coming from different range ambiguity regions in the joint transmit–receive domain, and further mitigated through minimum variance distortionless response criterion. In specific, to address the problems of insufficient snapshots and polluted samples enrolled in data-dependent beamforming, a robust beamforming approach utilizing the two-dimensional atomic-norm minimization is proposed to estimate the steering vector corresponding to the signal of interest and to reconstruct the interference-plus-noise covariance matrix. In this regard, to reduce the computational burden, the alternating direction method of multipliers is utilized to obtain both the true target's angle error and the jammer-plus-noise subspace. In addition, a series of theoretical proofs including equivalent semidefinite program transformation, and the optimality conditions are provided. The validity of the proposed mainlobe deceptive jammer mitigation approach is confirmed through numerical results.