When modelling temporal processes just like in pattern recognition, selecting the optimal number of inputs is a central concern. In this contribution, wetake advantage of specific features of temporal modelling to propose a novel method for extracting the inputs, that yield the best predictive performance.The method relies on the use of generalisation estimators to assess the performance of the model. This technique is first applied to time series processing,where we perform a number of experiments on synthetic data as well as a real life dataset, and compare the results to a benchmark physical method.Finally, the method is extended to system identification with a linear FIR filter.