This paper discusses a time-dependent classification approach for single trial recognition of spatial covert visual attention for Brain–Computer Interface. Covert visual attention is a natural and intuitive mental task that does not require any external stimulation. The possibility to recognize it from single trials is essential for a future online close-loop BCI. Experimental results indicate the feasibility of the proposed approach and its high performance in an offline study. We achieved an accuracy of 84.1±8.9% with a rejection of 6.5±6.6% averaged across subjects.