Abstract: When the spindle of NC machine tool is loaded, the micro vibration of the spindle under different conditions affects the machining accuracy, machining efficiency, tool wear and service life of the machine tool.In order to ensure the stability of machine tool spindle under load, a method of monitoring the vibration velocity of machine tool spindle and estimating the vibration by multivariate nonlinear fitting is proposed. Firstly, according to the three variable factors of machine tool spindle load - feed speed, spindle speed and cutting amount, the relationship between them and machine tool spindle speed vibration is established based on Levenberg Marquardt algorithm.Then, one of the three relational function models is selected as the basic function and associated according to “addition”, “subtraction”, “multiplication” and “division”. Finally, the combined function is used as the basic model for multivariate nonlinear fitting, the coefficient parameters are obtained, and the final fitting function is obtained. The experimental results show that this method can effectively predict the speed vibration of the spindle when the machine tool spindle is loaded, and the accuracy is more than 97%.