Canonical correlation analysis (CCA) is a promising feature extraction technique of steady state visual evoked potential (SSVEP)-based brain computer interface (BCI). Many researches have showed that CCA performs significantly better than the traditional methods. In this paper, the neural network implementation of CCA is used for the frequency detection and classification in SSVEP-based BCI. Results showed that the neural network implementation of CCA can achieve higher classification accuracy than the method of power spectral density analysis (PSDA), minimum energy combination (MEC) and similar performance to the standard CCA method. © 2013 Springer-Verlag Berlin Heidelberg.