Objective To investigate the pathogen composition and clinical features of preterm infants with sepsis, and to provide a basis for early identification and treatment of sepsis in preterm infants. Methods A retrospective analysis was performed for the clinical data of 371 preterm infants with sepsis who had a positive blood culture between January 2014 and May 2018. According to the time of onset, the preterm infants were divided into an early-onset group (an age of onset of < 7 days) with 73 preterm infants and a late-onset group (an age of onset of ≥ 7 days) with 298 preterm infants. The two groups were compared in terms of pathogen composition and clinical features (initial symptoms, laboratory examination results at the time of onset, comorbidities, and prognosis). Results There was a higher proportion of infants with Klebsiella pneumoniae infection in the late-onset group (P < 0.05), while there was a higher proportion of infants with Escherichia coli, Streptococcus agalactiae or Listeria infection in the early-onset group (P < 0.05). The early-onset group had a significantly higher proportion of infants with dyspnea than the late-onset group (P < 0.05). Compared with the late-onset group, the early-onset group had significantly shorter time to negative conversion of blood culture, duration of antibiotic use before infection, and indwelling time of deep venous catheterization (P < 0.05), and the late-onset group had a significantly higher incidence rate of neonatal necrotizing enterocolitis than the early-onset group (P < 0.05). The early-onset group had a significantly higher rate of treatment withdrawal than the late-onset group (P < 0.05). Conclusions Preterm infants with sepsis lack typical clinical manifestations and laboratory examination results at the time of onset. There are certain differences in pathogen composition and clinical features between preterm infants with early-and late-onset sepsis. Possible pathogens for sepsis should be considered based on age in days at the time of onset and related clinical features.