This paper presents a novel and simple encryption technique that provides very efficient and highly illusive encrypted data. A feedforward Neural Network (NN) that uses gradient descent methods is used in generating the encrypted data and the keys. The idea is about using the output of the hidden layer as the encrypted data. The resulting weights (hidden and output layers) are considered as the keys that are used for encryption/decryption at the sending and receiving ends. The input data is processed at the binary level format. Training is done for the neural network one time and the network is used for several transmissions of encrypted data until it is retrained for new keys. The training takes short time on multipurpose commercial computer to achieve very illusive encrypted data. Simulations and testing that have been achieved so far are encouraging for further future research. ICIC International © 2019.