We propose a new procedure for fast and robust extraction of the Steady-State Visual Evoked Potentials (SSVEP)for an asynchronous multi-command Brain Computer Interface(BCI). The SSVEPs were elicited by means of small reversal flickering checkerboards displayed on a monitor screen. In order to extract the SSVEPs we applied a new algorithm based on Constrained Independent Component Analysis with references (CICAr). A set of references defined as rectangular half-cycle-square periodic signals with the same frequencies as the frequencies of the flickering checkerboards were added into equality and inequality constraints, in order to extract in each moment the stronger elicited SSVEP activities. The final decision stage of the BCI system is performed by means of multiple statistical tests using surrogate data to contrast the reliability of the extracted SSVEPs by the CICAr algorithm, achieving a good discrimination when a subject is gazing or not gazing at a checkerboard. The proposed procedure does not require a classifier; therefore, it is not necessary to perform any calibration or training session to tune any classifier parameter, hence a plug and play BCI system. The algorithm has been tested with recorded EEG data of 4 healthy subjects from previously published real-time SSVEP-based BCI experiments. The obtained results confirm a high performance of the proposed algorithm, reaching a high transfer rate of information.