At the current stage, Brain-Computer Interfaces (BCIs) represent a promising technology for communication and control of assistive devices as well as for the clinical motor rehabilitation after a stroke. Current BCI systems may be divided in two main typologies, either based on the detection of the neural correlates elicited by predefined external stimuli or on the recognition of the self paced brain patterns related to mental imagination tasks. Nevertheless, in literature different observations have shown that not everyone can reach a good level of control with any of the current BCI typologies. In this regard, this thesis aims to show the suitability of a novel control signal for BCI: Covert Visuospatial Attention (CVSA). CVSA represents the ability of anyone to focus his attention at one point in space without the need of overt eye movements. This modality brings several advantages: first of all, it relies on the actual execution of an action (i.e., focusing attention) rather than on the pure task imagination. This represents an important benefit since it defines a control signal for BCI that it is natural, spontaneous and as close as possible to the daily experience of the user. Furthermore, it is totally gaze-independent and thus, it can be used by people with no (or limited) gaze control. Finally, CVSA does not require any external stimulation but can be exploited by a voluntary modulation of brain patterns. Recently, different studies started investigating CVSA to identify its neural correlates in scalp EEG. Nevertheless, no attempts of using CVSA for online EEG BCI operations are reported in literature. This thesis aims to demonstrate that an online EEG BCI based on CVSA is actually feasible and suitable. Therefore, three key aspects have been taken in account: (i) a new methodology for the detection and classification of CVSA from EEG signals, (ii) the evaluation of the online BCI operations by healthy users and, finally, (iii) the identification and the testing of two possible applications of CVSA BCI oriented to disabled users. The first part of this thesis is devoted to the definition of a new method for single trial classification of CVSA: the time-dependent approach. Previous studies have already demonstrated the involvement of the α-power over the parieto-occipital regions of the brain during CVSA tasks. The new method proposed here, extends these findings by means of more detailed analysis on the α sub-bands and on the temporal structure of the brain patterns. The intuition behind is twofold: on one hand, only specific bands are actually carrying discriminative information during the spatial attention tasks. On the other hand, attention-related patterns are evolving over time and consequentially a time-dependent approach would enhance the BCI performance. The analysis performed fully confirmed both the hypothesis and showed that the time-dependent approach can increase the classification accuracy of 12.3% (on average) with respect to classical methods used in literature. The aforementioned advances in the detection and single trial classification of CVSA allowed the implementation of the first online EEG BCI based on such a modality. Eight healthy subjects demonstrated the feasibility of online operating this BCI system. The performances achieved (above 70 %) and the general stability across days suggest that CVSA is a possible and promising candidate for controlling AT devices. On this goal, for the first time in literature, the CVSA BCI has been also investigated in more real conditions by introducing natural images in the protocol. Furthermore, the CVSA BCI has been evaluated by two end-users suffering from Amyotrophic Lateral Sclerosis disease as an alternative to standard BCIs. In particular, both of them have already tried a BCI system based on sensorimotor rhytms but with poor results. Conversely, with the CVSA BCI the best user reported an overall increment of 8.1 % compared to the previous modality and the remarkable level of 85 % of accuracy in the case of his best runs. This is the first attempt in literature to use CVSA as a BCI control modality with ALS users. Finally, this thesis proposes a novel and intriguing application in the field of cognitive rehabilitation. In particular, a pilot study has been conducted to assess the possible advantages of using the CVSA BCI during the treatment of the visual neglect syndrome. The hypothesis is that the integration of such a BCI—used as a reinforced feedback system—in addition with standard rehabilitation techniques might allow a deeper understanding of the recovery process as well as improving and speeding up the rehabilitative treatment. The results of this preliminary study suggest the validity of the hypothesis. The use of the online CVSA BCI seems to statistically influence the (re)activation of the affected hemisphere in both patients who participated in the treatment. This is also supported by evidences of a faster behavioral reaction time to the upcoming stimuli in the neglected vision hemifield. In conclusion, this thesis demonstrates the feasibility and the reliability of an EEG BCI based on the natural and spontaneous control signal of covert visuospatial attention. Beside the improvement in performance achieved by the newly proposed time-dependent approach, two exemplary applications have been identified and evaluated: (i) as an alternative to standard brain-computer interfaces for controlling AT devices and (ii) as a cognitive rehabilitation tool for visual neglect. In both the cases the results achieved open new perspectives and possibilities in the field of BCI research.