The　determination　of　an　informative　frequency　band　is　crucial　for　incipient　bearing　diagnosis.　Although　existing　methods　can　quantify　the　abundance　of　fault　information　within　candidate　improved　envelope　spectrums　(IESs),　the　selection　of　the　optimal　frequency　band　is　still　influenced　by　prior　knowledge.　To　tackle　this　issue,　an　adaptive　band　selection　method,　termed　IES　via　feature　significant　index　optimization-gram　(IESFSIOgram),　is　proposed　to　identify　the　optimal　frequency　band　for　bearing　fault　detection.　Firstly,　the　Spectral　coherence　(SCoh)　is　divided　into　several　candidate　bands　utilizing　a　1/3-binary　filter　bank.　Subsequently,　an　innovative　fusion　algorithm　is　devised,　which　deeply　integrates　the　adaptive　retrieval　capability　of　harmonic　product　spectrum　(HPS)　with　the　noise-resistant　property　of　diagnostic　feature　(DF)　to　calculate　the　feature　significance　index　(FSI)　and　ultimately　construct　the　IESFSIOgram.　The　effectiveness　of　the　method　is　verified　through　simulation　and　experiments,　demonstrating　its　robust　fault　identification　performance　on　some　occasions.　©　2024