A method and an apparatus for generating a secret key for encrypted communication using a synchronized neural network, which includes: generating initial codewords based on a bit string of weight values of the synchronized neural network and transmitting a first partial codeword of the initial codewords to a device of another party; receiving a second partial codeword generated by the device of the other party and combining final codewords based on the second partial codeword received from the device of the party and the bit string of the weight values; performing an error correction on the combined final codewords and transmitting first restoration success information according to the error correction to the device of the other party; and receiving second restoration success information from the device of the other party and generating the secret key based on the restoration success information of the device of the other party, are provided.