The shaft orbit is an important feature of rotating machinery. In orde r to research the shaft orbit for feature extraction and automatic identificatio n, three fractal dimensions, the box dimension, the information dimension and th e correlation dimension, were used to describe the fractal characteristics of th e shaft orbit. In addition, two geometrical characteristics as tightness and vol uptuous parameter were introduced into the analysis of the shaft orbit. The neur al network technology was used to classify the orbit shapes under different cond itions. The results show that by using the fractal and geometry characteristics of the shaft orbit,the application of neural networks to discriminating differen t operational states can get a high accuracy.