<i>Background:</i> Late-onset sepsis in the premature infant is frequently revealed by severe, unusual and recurrent bradycardias. In view of the high morbidity and mortality associated with infection, reliable markers are needed. <i>Objectives:</i> It was the aim of this study to determine if heart rate (HR) behavior may help the diagnosis of infection in premature infants with such cardiac decelerations. <i>Methods:</i> Electrocardiogram recordings were collected in 51 premature infants with a postmenstrual age <33 weeks with frequent bradycardias. Newborns in the sepsis group (C-reactive protein increase and positive blood culture) were compared with a no-sepsis group (C-reactive protein <5 mg/l before and 24 h after recording and negative blood cultures) for their HR characteristics, i.e. RR series distribution (mean, median, skewness, kurtosis, sample asymmetry), magnitude of variability in time and frequency domain, fractal exponents (α₁, α₂) and complexity measurements (approximate and sample entropy). Results are presented as the median (25%, 75%). <i>Results:</i> Gestational, chronological and postmenstrual age and gender were similar in the sepsis (n = 10) and no-sepsis group (n = 38). Three infants had an increase in C-reactive protein but negative cultures. Low entropy measurements [approximate entropy 0.4 (0.3, 0.5) vs. 0.8 (0.6, 1); p < 0.001] and long-range fractal exponent [α₂ 0.78 (0.71, 0.83) vs. 0.92 (0.8, 1.1); p < 0.05] were significantly associated with sepsis. No other HR characteristic was associated with sepsis. The decrease in 0.1 units of approximate entropy was associated with an over 2-fold increase in the odds of sepsis. <i>Conclusion:</i> Late-onset sepsis is associated with uncorrelated randomness of the HR. This abnormal HR behavior may help to monitor premature infants presenting with frequent and severe bradycardias.