Résumé : Background. No sensitive, specific marker able to discriminate favourable or unfavourable outcome of fever of unknown origin (FUO) at diagnosis has been identified. Procalcitonin, a recently assessed infection marker, may be useful in predicting the outcome of FUO. Methods. We conducted a prospective study examining the following variables: age 0.5-22 years; solid tumour diagnosis; chemotherapy-related grade-4 febrile neutropenia (FN). A complete clinical, bacteriological and biological evaluation was performed at hospital admission (H0). Other investigations depended on clinical status. FUO was considered to be of unfavourable outcome if the fever was persistent or re-appeared at day 3 (or later), or if secondary clinical or microbiological infection occurred. To validate the results of the analysis the data set was randomly split into a training set and a validation set. Results. Out of 172 episodes of FN, 136 episodes were classified as FUO (80%). Seventy-two (53%) were included in this study. PCT values were significantly higher in episodes of unfavourable outcome (P<0.001). None of the other prediction candidates appeared to be significantly linked to the risk of unfavourable outcome. In the validation set, the best PCT cut-off was 0.12 μ/L, which was associated with a sensitivity of 80% and specificity of 64%. Conclusions. PCT-H0 level can predict FUO outcome. A protocol based on PCT-H0 measurement, integrating clinical and bacteriological evaluation, facilitates shorter hospital stays and less antibiotic treatment. Patients with a PCT-H0 value <0.12 μ/L could benefit from an outpatient treatment starting at H48 thus reducing hospitalisation costs and improving quality of life. © 2009 Wiley-Liss, Inc.