Recently, several studies have started to explore covert visuospatial attention as a control signal for brain-computer interfaces (BCIs). Covert visuospatial attention represents the ability to change the focus of attention from one point in the space without overt eye movements. Nevertheless, the full potential and possible applications of this paradigm remain relatively unexplored. Voluntary covert visuospatial attention might allow a more natural and intuitive interaction with real environments as neither stimulation nor gazing is required. In order to identify brain correlates of covert visuospatial attention, classical approaches usually rely on the whole α-band over long time intervals. In this work, we propose a more detailed analysis in the frequency and time domains to enhance classification performance. In particular, we investigate the contribution of α sub-bands and the role of time intervals in carrying information about visual attention. Previous neurophysiological studies have already highlighted the role of temporal dynamics in attention mechanisms. However, these important aspects are not yet exploited in BCI. In this work, we studied different methods that explicitly cope with the natural brain dynamics during visuospatial attention tasks in order to enhance BCI robustness and classification performances. Results with ten healthy subjects demonstrate that our approach identifies spectro-temporal patterns that outperform the state-of-the-art classification method. On average, our time-dependent classification reaches 0.74 ± 0.03 of the area under the ROC (receiver operating characteristic) curve (AUC) value with an increase of 12.3% with respect to standard methods (0.65 ± 0.4). In addition, the proposed approach allows faster classification (<1 instead of 3 s), without compromising performances. Finally, our analysis highlights the fact that discriminant patterns are not stable for the whole trial period but are changing over short time intervals. These results support the hypothesis that visual attention information is actually indexed by subject-specific α sub-bands and is time dependent.