Biometric authentication involves various technologies to identify individuals by exploiting their unique, measurable physiological and behavioral characteristics. However, biometric authentication systems (e.g., face recognition, iris, retina, voice, and fingerprint) are increasingly facing the risk of being tricked by biometric tools such as anti-surveillance masks, contact lenses, vocoders, or fingerprint films. In this regard, we design a multimodal biometric authentication system named DeepKey which uses both gait and Electroencephalography (EEG) signals to provide better protection against such risks. DeepKey consists of three key components: an Invalid ID Filter Model to block invalid subjects, a Gait Identification Model to recognize Gait IDs and an EEG Identification Model to recognize EEG IDs. In particular, the first two models employ a one-class SVM algorithm and a Recurrent Neural Network based deep learning model, respectively. The third model combines autoregressive coefficients, an RNN structure, and an SVM classifier. DeepKey is trained with a gait dataset of 160,000 samples and an EEG dataset of 108,000 samples. Experimental results show DeepKey outperforms a series of comparison methods and achieves an overall accuracy of 0.983 along with an overall false acceptance rate (FAR) of 0.0 and a false rejection rate (FRR) of 0.019.