The principal aim of this report is to present the results of multivariate analyses conducted to identify clinical prognostic factors in 92 children aged < 16 years with ependymoma (EPD) retrospectively collected in seven Italian centres. They were treated over a 16-year period (1977-1993). Treatment modalities varied. Surgery and radiotherapy (RT) was the "gold standard" management method for the majority of these children. Only in the late 1980s did some of them receive chemotherapy (CT), mainly with vincristine, lomustine (CCNU) and prednisone. The median follow-up of the entire study population is 36 months (average 43 months; range 12 to 214 months). The 10-year overall (OS) and the progression-free survival (PFS) of the study population were 55.5% (CI 41.4-69.4%) and 34.7% (CI 21.4-47.8%), respectively. Age (< 5 years; > 5 years), sex, site (infratentorial vs. supratentorial), histology (anaplastic/malignant vs. non-anaplastic/non-malignant), type of resection (complete vs. incomplete); use and fields of RT, and of CT employed were entered in a multivariate regression model to test their impact on OS and PFS. On univariate analysis, radical surgery, the use of RT and age more than 5 years at the time of diagnosis achieved statistically significant values for predicting long-term OS and PFS. Histology reached marginal statistical significance but only for PFS. When those variables were entered in a multivariate analysis only radical resection (P = 0.00142 and 0.0001) resulted a significant factor for predicting long-term OS and PFS, while the use of RT reached a marginal statistical significance, but only for PFS (P = 0.05). Children who had the tumour completely resected did significantly better than all the others who had less than a complete resection, with a 10-year OS and PFS for the two groups of patients of 69.8% (CI 53-86.5%) and 57.2% (CI 40.3-75%) and of 32.5% (CI 8.5-57.6%) and 11.1% (0-24.4%), respectively. These findings suggest that, for childhood EPD, radical resection should be pursued as much as reasonably possible. Thus, it seems justified proposing for future trials, patient stratification by entity of surgical resection.