The small load omitting approach is a branch of accelerated fatigue test technology. The existing methods hardly ever consider the randomness of the fatigue property of the structure. In this paper， the fatigue life is regarded as a continuous random variable， and the difference of fatigue life distribution before and after small-load-omitting is quantitatively described by probabilistic distance. The standard， in which the distribution of the fatigue life of the omitted spectra is the same with that of the original spectra， is used. The dispersion of fatigue property of structure is considered in this approach. According to the distribution density function of stress amplitude and the p-S-N curves of the component， the omission level can be calculated. Fatigue tests with notched specimens made of LC4CS aluminum alloy are conducted to verify this approach. Five different spectra are obtained by removing small cycles of which amplitudes are less than the specified omission level， and the number of load cycles is about 8.4% of the original spectra. It is found that experimental results agree well with theoretical solutions.