In the use of large computers to analyze severe accidents in liquid-metal fast breeder reactors (LMFBRs), it has long been recognized that many of the fundamental phenomena cannot be precisely predicted because of uncertainty in the parameters that govern them. As a direct result, mechanistic analysis of such accidents has proceeded along a parametric path in which these variables are fixed at a certain constant value for the entire calculation: The influence of variation of this value is assessed by making a series of complete calculations with the parameter set at a different value for each such element of the series. While some parameters may be thought of as “correlated” or fixed for an entire calculation, very few are in fact constant throughout a reactor, and many are (for practical purposes) nearly completely uncorrected, either in space or time, during the hypothetical accident. Thus, such analysis has created a set of results that are not indicative or representative of an accident involving uncorrected or only partially correlated variable parameters. We describe here a methodology for dealing with various degrees of uncertainty or incoherence in these parameters. By using two very different mechanistic codes (FX2-POOL and EPIC), we demonstrate that the treatment of uncorrected parameters, such as droplet/particle size in a hypothetical core disruptive accident, as random variables with a certain probability distribution during each complete calculation of a series of calculations produces as much as an order of magnitude less uncertainty in the end result than had been obtained assuming perfect correlation. Finally, we categorize a small list of parameters as either correlated or uncorrected for some of the other LMFBR accident analysis codes. The technique we demonstrate can be easily implemented in a broad spectrum of accident analysis codes with similar benefits.