Indirect, online tool wear monitoring is one of the most difficult tasks in the context of industrial machining operation. The challenge is how to construct an effective model that can consistently exemplify the degradation propagation of tool performance (i.e., tool wear) based on a continuous acquisition of multiple sensor signals. This paper proposes an adaptive Gaussian mixture model (AGMM) to provide a comprehensible and robust indication (i.e., Kullback–Leibler (KL) divergence) for quantifying tool performance degradation. Based on dynamic learning rate, parameter updating, and merge and split of Gaussian components, AGMM is capable of online adaptively learning the dynamic changes of tool performance in its full life. Furthermore, the performance changes of tools are quantified by measuring the distance between two density distributions approximated by the AGMM and the baseline GMM trained by the normal data, respectively. Experimental results of its application in a machine tool test demonstrate the effectiveness of the AGMM-based KL-divergence indication for assessment of tool performance degradation.