A systematic approach to statistical simulation for large scale analog circuits is presented. The statistical model takes into account mismatch between devices due to variations in the process and noise, as well as interdie variations. The number of statistical parameters is O (n), where n is the number of devices. Additionally, because large-scale analog circuits are very computationally intensive to simulate, a two level approach is used, combining a behavioral model relating block performances to circuit performances and regression models relating a set of primary statistical variables to block performances. The efficiency of this approach to statistical simulation is demonstrated with the examples of an A/D converter.