Pulse wave contains a lot of cardiovascular diseases (CVD) information. It can be used as an early diagnostic method for CVD. Previous studies have shown that the pulse wave of atherosclerotic patients are significantly different from those of healthy subjects. In this study, we extracted 210 pulse wave cycles from atherosclerotic patients. At the same time, as a control group, we extracted the same number of pulse cycles from healthy subjects. An optimized convolution neural network (CNN) was proposed to classify these two pulse patterns. The proposed CNN had good performance with 95% accuracy on distinguishing arteriosclerosis vs non- arteriosclerosis. Our study showed that CNN could identify pulse waves in atherosclerosis patients with high accuracy. It could help to develop a non-invasive and efficient method for arteriosclerosis detection.