A single hidden layer feedforward neural network (FFNN) is called a transparent FFNN if its output vector is a reproduction of the input vector. In this case, the input vector is the target output vector. Once the transparent FFNN is well designed with the m-bit plaintext input set, the designed network is divided into two parts: the transmitter private network encrypting part and the receiver public network decrypting part. The hidden vector which is the output real vector of the private network part is transmitted using the decimal-value ordered lookup table (DVOLT). The m-bit binary cipher vector is the chosen binary index of the lookup table. In this paper, the binary plaintext vector size (m) is chosen to be 16 bits (two bytes). Computer simulation shows that both operations of encryption or decryption of all possible plaintext of 65536 (216) vectors can be done with zero error and an average processing time of 8.3 msec per 16-bit vector (4.15 msec / byte) per operation. The average hamming distance between the binary plaintext and the binary ciphertext is calculated as 7.8798 for all possible plaintext of 65536 16-bit vectors. This scheme can't be attacked by either brute force or cryptanalysis since there are no binary keys or known mathematical structures. The most important part which must be kept secret is the private matrix of the transmitter private network encrypting part. Each of the private encrypting key and the public decrypting key needs a memory size of about 8.5 Mbytes. The large size of both the private and public keys may limit the applications of the proposed scheme to large workstations intercommunications. This proposed cryptography scheme provides sender authentication and also receiver confidentiality.