Nuclear criticality safety analysis using computational methods such as a Monte Carlo method must establish, for a defined area of applicability, an upper subcritical limit (USL), which is a calculated multiplication factor k that can be treated as actually subcritical and is derived from a calculational margin (combination of bias and bias uncertainty) and a margin of subcriticality. Whisper, a nonparametric, extreme-value method based on sensitivity/uncertainty techniques and the associated software are presented. Whisper uses benchmark critical experiments, nuclear data sensitivities from the continuous-energy Monte Carlo transport software MCNP, and nuclear covariance data to set a baseline USL. Comparisons with a traditional parametric approach for validation, which requires benchmark data to be normally distributed, show that Whisper typically obtains similar or more conservative calculational margins; comparisons with a rank-order nonparametric approach show that Whisper obtains less stringent calculational margins.