It is very difficult to avoid the problem of tool wear or breakage during the machining of CNC machine tools. Failure to detect and stop the machine in time is likely to cause damage to the machine or workpiece. This requires the on-line monitoring of the tool state during machining. In this paper, a load-based tool wear monitoring method is proposed. The self-learning method is adopted to obtain the load range of a specific machining process, so that on-line monitoring of the subsequent same machining process can be carried out. Self-learning is mainly used to process the load signal by using the statistical algorithm (6σ algorithm), so as to obtain the upper and lower boundary of the monitoring range. The load signals in the subsequent machining process are compared with the upper and lower boundaries determined by the self-learning algorithm to determine whether the tool is worn or not. Finally, the feasibility of this method was verified by experiments.