Infections are a major cause of death in neonates. Diagnosis of neonatal sepsis is a major challenge because newborns have very nonspecific clinical signs and auxiliary tests have low sensitivity. In order to improve the correct diagnosis of this condition, we propose an algorithm of diagnostic surveillance for late neonatal sepsis in Peru and countries of the region. The algorithm classifies the episodes as confirmed, probable or possible sepsis, and especially seeks to identify those episodes that do not correspond to sepsis, preventing other diseases to be qualified as "sepsis". Better diagnostics will enable more realistic rates of neonatal sepsis, improve the use of antibiotics and avoid their negative effects on newborns, as well as provide a more accurate view of their impact on public health.