Recently, a new spatiotemporal filter based on linearly constrained minimum variance (LCMV) beamforming was introduced for steady-state visual evoked potentials (SSVEPs) detection. Due to the use of calibration data and the time-domain approach, the filter can be optimized for each individual participant and significantly improve the signal to noise ratio. Therefore, spatiotemporal beamforming (STBF) can achieve higher classification accuracy compared with other widely used canonical correlation analysis (CCA)-based methods. In this study, we propose a novel method that combines a filter bank approach with STBF, to learn discriminative features from fundamental and harmonic frequency components of SSVEPs and further improve the frequency detection rate. Using 12-class SSVEP datasets recorded from ten participants, we compared the performance of our new detection method with that of four other methods, CCA, IT-CCA, Comb-CCA and STBF. The obtained results show that our FB-STBF method can achieve 85.67% classification accuracy, which is higher than all other approaches. Thus, we suggest that our novel FB-STBF method is suitable for implementing high-performance SSVEP-based BCI systems