To distinguish children with chemotherapy-induced febrile neutropenia (FN) at low risk of severe infection, the variables that are significant risk factors must be identified. Our objective was to identify them by applying evidence-based standards. This retrospective 2-center cohort study included all episodes of chemotherapy-induced FN in children in 2005 and 2006. The medical history, clinical, and laboratory data available at admission were collected. Severe infection was defined by bacteremia, a positive culture of a normally sterile body fluid, invasive fungal infection, or localized infection at high risk of extension. Univariate analysis identified potential predictive variables. A generalized mixed model was used to determine the adjusted variables that predict severe infection. We analyzed 372 FN episodes. Severe infections occurred in 16.1% of them. Variables predictive of severe infection at admission were: disease with high risk of prolonged neutropenia (adjusted odds ratio [aOR]=2.5), blood cancer (aOR=1.9), fever ≥38.5°C (aOR=3.7), and C-reactive protein level ≥90 mg/L (aOR=4.5). Now that we have identified these variables significantly associated with the risk of severe infection, they must be validated prospectively before combining the best predictive variables in a decision rule that can be used to distinguish children at low risk.