While CNC machining, most defects of machining accuracy are caused by aging or wearing of spindle bearing. It is not practical to calculate vibration features from specific frequency band to detect the status of CNC spindle. It’s hard to define all type of defect. Also, all the preload, lubricating, maintaining and the attachment of sensor will affect the frequency response. Users can only make sure the machine/equipment is health or sub-health. Therefore, a health modeling which is based on the “health data” is more practical. We have proposed and developed a spindle health diagnosis system which features cost-effectiveness, highly-efficiency and module-design. With collecting the normal operation data while machine booting, the system can build the diagnosis model by tuning optimal parameters automatically and provide the probability of varied parameter when detecting the abnormal status.