In nanoscale integrated circuit technologies, process parameter fluctuations gain increasingly in importance. Efficient methods are thus required during the design phase to evaluate the resulting variability. In this letter, we propose a new method to estimate the variation bounds of analog circuit performance. This method combines design of experiment techniques with the Cornish-Fisher expansion: process parameter variations are first mapped to circuit performance metrics by a quadratic model, and then an analytical approximation of the performance distribution's quantiles enables the enclosure of the performance variations. The proposed method demonstrates a better accuracy/efficiency ratio than Monte-Carlo-based methods.