A novel fault tracking approximator for fault diagnosis scheme of high voltage direct current transmission（HVDC） system is proposed. A consensus filter is used to mitigate the random noise at the d.c. side of HVDC. According to the HVDC state space equation， a new parameter， i.e. virtual fault， is introduced to represent the actual fault to construct the fault tracking approximator. Based on the predictive theory， an optimization time span is then chosen. Using the iterative algorithm， virtual fault is adjusted to approximate actual one in terms of the difference between outputs of the fault tracking approximator and the actual system. Different from those traditional fault diagnostic methods grounded on neural networks or support vector machines， the algorithm proposed can not only detect the system faults， estimate the fault function， but diagnose different sorts of faults. Finally， simulation results show the feasibility and effectiveness of the approach proposed.