A method for determining the proper regression vector for recreating the dynamics of nonlinear systems is presented The false nearest neighbors (FNN) algorithm, originally developed to study chaotic time series, is used to determine the proper regression vector for input/output system identification and inferential prediction using only time-series data. The FNN algorithm for solving these problems is presented, and the problem of analyzing noise corrupted time series is discussed. The application of the algorithm to a number of examples including an electrical-leg stimulation experiment, an industrial pulp-digester model, a polymerization model, and a distillation-column simulation is presented and the results are analyzed.