Response surface techniques are presented for obtaining the probability distributions of selected consequences of a liquid-metal fast breeder reactor hypothetical core disruptive accident. The uncertainties of the consequences are considered as a variability of the system and model input parameters used in the accident analysis. Probability distributions are assigned to the input parameters, and parameter values are systematically chosen from these distributions. These input parameters are then used in deterministic consequence analyses that are performed by fast-running analogs of the comprehensive mechanistic accident analysis codes. The results of these deterministic consequence analyses are used to generate the coefficients for response surface functions that approximate the consequences in terms of the selected input parameters. These approximating functions are then used to generate the probability distributions of the consequences with random sampling being used to obtain values for the accident parameters from their distributions. Two different schemes are presented for selecting the knot-point values of the input parameters. The first generates a single second-order polynomial for the entire parameter space; the second generates separate polynomials for specified regions of the parameter space. A technique to handle nonindependent or correlated input parameters is presented. Finally, the calculation of conditional distributions of the consequences and the use of these distributions to define importance distributions of the input parameters are presented. The use of these procedures is illustrated by applications to a postulated loss-of-flow transient with failure to scram in a Clinch River Breeder-type reactor.