The classical techniques for fault diagnosis require periodic shut down of machines for manual inspection. Although these techniques can be used for fault diagnosis in simple machines, they can rarely be used effectively for complex ones. Due to the rapid growing market competitiveness, more reliable and robust condition monitoring systems are critically needed in a wide array of industries to improve production quality and reduce cost. As a result, in recent years more efforts have been taken to develop intelligent techniques for online condition monitoring in machinery systems. Several neural fuzzy classification schemes have been proposed in literature for fault detection. However, the reasoning architecture of the classical neural fuzzy classifiers remains fixed, allowing only the system parameters to be updated in pattern classification operations. To improve the reliability of machinery fault diagnostics, an evolving fuzzy classifier is developed in this work for gear system condition monitoring. The evolution is performed based on the comparison of the potential of the incoming data set and the existing cluster centers. One key feature of the developed evolving fuzzy classifier is that it has the ability of developing continuously - by adding or subtracting rules and by modifying existing rules and parameters. In performance evaluation, the proposed evolving classifier is firstly tested with the use of benchmark data sets, such as Iris data, Wisconsin breast cancer data and wine data. Then the adopted evolving classifier is implemented for gear fault diagnosis. A distinguishable pattern is determined between the input data and the output patterns to evaluate the data sets. Several signal processing techniques are utilized to generate representative features to train the proposed evolving fuzzy classifier. Simulation test results show that the proposed classifier can effectively identify the condition of a gear, both spur and helical types, and it outperforms provide other related methods.