In this paper, we present a predictive neural network called Neural Predictive Coding (NPC). This model is used for non linear discriminant features extraction (DFE) applied to phoneme recognition. We validate the nonlinear prediction improvement of the NPC model. We also, present a new extension of the NPC model : NPCJ. In order to evaluate the performances of the NPC-3 model, we carried out a study of Darpa-Timit phonemes (in particular /bl, Id/, lgl and /p/, /U, /q/ phonemes) recognition. Comparisons with traditionnal coding methods are presented: they put in obsviousness an improvement of the classification. We also show how an adaptative constraint allows improvements on recognition task 1 0-7803-7278-6/02/$10.00 02002 EEE