The practical use of speech recognition systems is often restricted by the influence of the acoustical environment. Usually speech recognizers are developed in laboratories which do not reflect the practical working conditions such as additive back-ground noise as e.g. inside a car. Various studies (1),(2) have already shown that noise signals have a considerable influence on the recognition rate. In this contribution firstly the influence of noise on a speaker independent isolated word recognizer will be described. Then several possibilities to improve the recognition by the use of a noise suppression method are shown. The noise suppression is based on a spectral subtraction technique.