Objective To analyze the distribution, changes and drug resistance of pathogenic bacteria in late-onset sepsis (LOS) in neonatal intensive care unit (NICU). Methods The clinical data of 223 LOS neonates with positive blood culture admitted to NICU from January 2012 to December 2019 were retrospectively analyzed, and the changes of pathogenic bacteria from year to year were analyzed. Results Among 223 neonates (116 boys and 107 girls), the mean gestational age was (31.9±1.2) weeks, the mean birth weight was (1584.1± 620.9) g, and the median age of LOS was 19.0 (13.0-27.0) days. There were 203 preterm infants and 20 full-term infants. A total of 234 pathogenic strains were detected, most of which were G^- bacteria (128 strains, 54.7%), including 77 strains of Klebsiella pneumoniae. Sixty-nine strains (29.5%) were G⁺ bacteria and 37 strains (15.8%) were fungi. There were significant differences in the detection rates of G^- bacteria, G⁺ bacteria and fungi among different years (P<0.01). The highest detection rates of G^- bacteria and fungi were found in 2016, and the highest detection rates of G⁺ bacteria were found in 2014. The onset time, cesarean section rate, time of premature rupture of membranes >18 hours, antibiotic exposure rate and mortality among G⁺ bacteria, G^- bacteria and fungi groups were significantly different (P<0.05). The cesarean section rate and mortality were higher, and the onset time was shorter in G^- bacteria group. The rate of antibiotic exposure and the proportion of premature rupture time >18 hours were higher in fungi group. Klebsiella pneumoniae (34.9%) was the commonest pathogen causing LOS in premature infants. There was no significant difference in the detection rates of G^- bacteria, G⁺ bacteria and fungi between premature and full-term infants (P>0.05). A total of 109 strains (46.6%) of multiple drug-resistant bacteria were detected and extended-spectrum β-lactamase (ESBLs)-producing Enterobacteriaceae was the dominant strain (74 strains, 67.9%). Conclusions G^- bacteria are the common pathogens causing LOS in NICU, and have serious resistance to commonly used antibiotics. Retrospective analysis of pathogen distribution and resistance characteristics in the NICU on a regular basis is helpful to guide the rational and effective use of antibiotics.