In this paper, we investigate the robust beamforming for the magnetic-based wireless power transfer (WPT) with multiple transmitters and multiple receivers to improve the power efficiency and reliability. By taking into account the imperfection of the magnetic mutual inductances information (MII), we firstly formulate a beamforming optimization problem to minimize the total transmit power while guaranteeing the reliable power delivery for each receiver simultaneously. Then, observing that the optimization problem is highly nonconvex with infinite number of constraints, we utilize the norm-bounded MII error models to reformulate the original problem into two worst-case optimization problems for the cases of the single receiver and multiple receivers, respectively. Finally, the optimal beamforming is obtained for the case of single receiver by applying the semidefinite programming relaxation (SDR) and the approximately optimal solution is obtained for the case of multiple receivers by adopting the SDR with randomization. Extensive simulation results verify the power transfer reliability and efficiency of the proposed beamforming schemes for the magnetic-based WPT systems.