This paper introduces a new algorithm for the early detection of abnormal arcing conditions of circuit breakers (CBs). Any interruption with a higher arcing time compared to maximum permissible arcing time of CB is defined as a CB failure to clear mode. Measuring CB arcing time is not easy because there is no control on the mechanical opening and closing instants of CB contacts. Nevertheless, there is a correlation between the spectral energy densities of CB arcing voltages and arcing times of CB. This correlation is used by calculating the spectral energy densities of the instantaneous system voltage and the instantaneous CB voltage across its terminals utilizing discrete Fourier transform for two periods of the system frequency. The difference between the spectral energy densities at the fundamental frequency is defined as a criterion for identifying a failure to clear mode. Whenever this criterion exceeds a predefined trip level, it indicates the occurrence of a failure to clear mode. In this situation, a trip signal is initiated for adjacent CBs to isolate the faulted CB and the fault. The results obtained from computer simulations and measurements show that the algorithm discriminates between normal and abnormal arcing conditions of CB.