The waveform characteristics of pulse signal can be used to diagnose early atherosclerosis. Based on the research achievement of predecessors, eight feature parameters including both time-domain and frequency-domain related to arteriosclerosis are chosen for research so that we can analyze the pulse signal in many aspects. Using probabilistic neural network to train model of arteriosclerosis, spread constant of neural network is optimized through the adaptive genetic algorithm. Through the simulation experiment, it shows that probability neural network can predict the arteriosclerosis well. A feasible method which can achieve the classification of arteriosclerosis is provided.