In this paper, we propose two secure multiuser multiple-input multiple-output transmission approaches based on interference alignment (IA) in the presence of an eavesdropper. To deal with the information leakage to the eavesdropper as well as the interference signals from undesired transmitters (Txs) at desired receivers (Rxs), our approaches aim to design the transmit precoding and receive subspace matrices to minimize both the total inter-main-link interference and the wiretapped signals (WSs). The first proposed IA scheme focuses on aligning the WSs into proper subspaces while the second one imposes a new structure on the precoding matrices to force the WSs to zero. When the channel state information is perfectly known at all Txs, in each proposed IA scheme, the precoding matrices at Txs and the receive subspaces at Rxs or the eavesdropper are alternatively selected to minimize the cost function of an convex optimization problem for every iteration. We provide the feasible conditions and the proofs of convergence for both IA approaches. The simulation results indicate that our two IA approaches outperform the conventional IA algorithm in terms of average secrecy sum rate.