In this paper, we propose an efficient code-book-based symbolwise beamforming for millimeter-waveWPAN systems, which is based on the multi-level training and antenna selection to reduce the protocol overhead in terms of the number of required training sequences. Using our proposed antenna selection method, one beam direction of the specific level includes two beam directions of the following level, which results that the required number of training sequences at each level is constant regardless of the number of antennas. Once antennas of the transmitter and receiver are selected properly at each level, then the training sequences are exchanged with the direction specified by the pre-defined codebook. By adopting the proposed antenna selection method and multi-level training, our proposed scheme can significantly reduce the total number of required training sequences for the beamforming setup. To verify the performance, we compare our proposed scheme with the conventional symbol-wise beamforming schemes based on the codebook. It is evident from the simulations that our proposed scheme can provide the effective signal-to-noise ratio (SNR) gain approaching to the conventional scheme with fewer training sequences.