On-line detection system of the abnormal states in a machining process needs to be developed to implement the IMS(Intelligent Manufacturing System). High productivity and efficient quality control can be achieved through the on-condition maintenance for normal tool condition. Generally it is difficult to determine the exact point of time for a tool change because a tool wear grows gradually on the contrary to other abnormal states such as tool fracture, chattering etc. In this article, the shape variation of cutting force signal generated by a insert during face milling was investigated along with a tool wear. The variance, skewness and kurtosis were used as the shape parameters to describe the shape variation and, consequently, utilized as the features to monitor a tool wear. Experimental results showed that the shape parameters could discriminate the tool condition reliably between a fresh tool and a worn tool. As a result, we proposed the method to diagnose a tool wear by combining these parameters with a neural network algorithm.