This paper presents an algorithm for transmission line fault diagnosis and classification using K-means clustering. The three-phase current signal is synchronized with the GPS clock and approximate wavelet coefficients are computed over a moving window of one cycle. Two centroids of successive cycles are computed using K-means clustering. The centroidal difference of local and remote bus are added to obtain resultant centroidal difference. The resultant centroidal difference is compared with the threshold to detect the fault. Similarly, the centroidal difference is computed of three-phases to classify the fault. A number of case studies have carried out to validate the proposed algorithm by fault impedance, fault incidence angle and, fault location. The robustness of the algorithm has been established in the presence of noise. © 2018 IEEE.