In this paper we describe a speech recognition system using linear dynamic models and articulatory features. Experiments are reported in which measured articulation from the MOCHA corpus has been used, along with those where the articulatory parameters are estimated from the speech signal using a recurrent neural network.