In order to reliably monitor unexpected tool failure and prevent workpiece or machine tool from possible damages in batch machining, a tool breakage on-line monitoring method based on power information and cross-correlation algorithm is proposed. In this method, wavelet coefficients of spindle-power signal are used as the characteristic vector of machining information, and then the vector sequence extracted from a normal machining process via Mallat wavelet is defined as the reference template for monitoring cutting tool condition. In batch machining, real-time characteristic vector of the workpiece in machining process is extracted via an improved real-time wavelet algorithm. The correlation of two vector sub-sequences within a sampling time window, which is described by generalized cross-correlation coefficient, decreases apparently when the tool is broken. The generalized cross-correlation coefficient is defined as tool condition index (TCI), and tool breakage can be detected by monitoring the TCI with a threshold value. Experiments show that the method can accurately identify tool breakage failures in normal machining condition, and thus it is practical.