IBM's Underwriting Profitability Analysis application mines property and casualty (P&C) insurance policy and claims data to construct predictive models for insurance risks. UPA uses the ProbE (probabilistic estimation) predictive-modeling data-mining kernel to discover risk-characterization rules by analyzing large and noisy data sets. Each rule defines a distinct risk group and its risk level. To satisfy regulatory constraints, the risk groups are mutually exclusive and exhaustive. ProbE generates rules that are statistically rigorous, interpretable, and actuarially credible. The authors validated this approach in a joint development project with a P&C firm. The results suggest that this methodology provides significant value to P&C insurance risk management.