The improved envelope spectrum via feature optimization-gram (IESFOgram) is an effective tool, which can determine an appropriate demodulation band for improving the performance of the cyclic spectral coherence algorithm in revealing the fault feature of the rolling element bearing (REB), in which a diagnostic feature (DF) indicator is employed to assess the abundance of bearing fault information in an each divided band. However, the difference between the theoretical and actual characteristic frequencies of the REB is not considered in the DF indicator, which may make the determined demodulation band invalid for revealing fault characteristics on some occasions. To address this issue, an improved diagnostic feature (IDF) based on the frequency tolerance and soft threshold denoising tool is proposed in this study. Firstly, the interference from the difference between the theoretical and actual characteristic frequencies of the REB on the DF indicator is eliminated by setting the frequency tolerance. Secondly, the robustness of the IDF indicator is further improved by the adaptive soft threshold denoising technology of each harmonic order. The effectiveness of the proposed scheme is verified by simulation and experiments, and the results show that the proposed scheme has better performance to reveal bearing fault characteristics on some occasions.