Recently, Yen and Guo proposed a chaotic neural network (CNN) for signal encryption, which was suggested as a solution for protection of digital images and videos. The present paper evaluates the security of this CNN-based encryption scheme, and points out that it is not secure from the cryptographical point of view: 1) it can be easily broken by known/chosen-plaintext attacks; 2) its security against the brute-force attack was much over-estimated. Some experiments are shown to support the results given in this paper. It is also discussed how to improve the encryption scheme.