Development of input-output models for nonlinear systems have gained attention recently. A knowledge-based system (KBS) is being developed for constructing input-output models of nonlinear dynamic processes. The KBS automates outlier detection and triggers the execution of advanced nonparametric modeling techniques, such as parsimonious polynomial approximation and multivariable adaptive regression splines. The software combines heuristic search methods and reasoning ability of the KBS with statistical inferences to detect outliers, determine the nonlinearity of the system, identify the nonlinear or linear models and validate them automatically.