Abstract: Chaotic phase state classification is a key step in utilizing chaotic systems to detect weak signals. A chaotic phase state classification method based on an improved pulse coupled neural network (PCNN) was proposed. The movement characteristics of mammalian visual cortical neurons were simulated with improved PCNN, while outputting the structural characteristics of a phase state diagram of a chaotic system. Dimensions of these characteristics were reduced using average residual error to yield real-time classification for chaotic/periodic states of the system. Sinusoidal and ECG signals were used to verify the proposed method using Lyapunov characteristic exponents. Results indicate that the method can quickly and accurately classify different chaotic phase states.