The main problem encountered when applying remote sensing and geographic information systems techniques for wildfire risk assessment is the necessity to integrate different data sources. The methods applied so far are usually based on regression techniques or on coefficients relying on experts' knowledge. Hence fire managers are seeking an unbiased statistical model able to highlight the multivariate spatial relationships between the predictor variables, yielding understandable output readily accessible to end users. The present research aims to test the capability of classification and regression trees (CART) analysis to assess long-term fire risk at a local scale. The CART analysis is a nonparametric statistical technique which generates decision rules in the form of a binary tree, for a classification or a regression process. A fire-prone study area was selected in the southeast of Italy. Fire ignition points, relative to a 7 year period (1997--2003), were used to derive a fire occurrence map through a kernel density approach. The resulting map was then used as input response variable for the CART analysis with fire danger variables used as predictors. The rules induced by the regression process allowed the definition of different risk levels, expressed as 30 management units, which is useful for producing a fire risk map. The result of the regression process (r = 0.77), the capability of the CART analysis to highlight the hierarchical relationships among the predictor variables, and the improved interpretability of the regression rules represent a possible tool useful for better approaching the problem of assessing and representing fire risk.