A study was carried out to compare the effects of telephone transmissions on different speaker-independent isolated-word recognition systems. Two databases of telephone-quality utterances were recorded from 150 speakers, the first over the Paris analog network, and the second over a local private network. The recordings covered a wide range of speakers, background noise environments and telephone transmission conditions. Data sets made up from these telephone-quality utterances were then used to evaluate and compare recognition algorithms based on, firstly, dynamic time warping and, secondly, Markov modelling. Recognition tests performed using French digits confirmed that the use of telephone-quality speech degrades the recognition performance. Using a dynamic time warping algorithm, recognition rates were obtained of 70% over the analog network and 82% over the local private network. A slightly better performance was obtained using Markov modelling under the same conditions, the figures obtained here being 85% and 90% respectively.