The main goal of this work is to evaluate a given implementation of a detector of landmarks. In this context, /andmarks are positions in speech signals, where much information about the activities of articulators, which produced the signal, can be found. This information leads to distinctive features, which help to achieve the spoken speech sounds. The detector checks speech signals for three types of landmarks: glottal, sonorant and burst. Expected landmarks are needed for being able to test the success of the detector. So first of all, these expected landmarks have to be determined. Therefore, a corpus, which contains read speech and spontaneous speech with labelled phonetic information, is used to extract these landmarks. This extraction is one of the main jobs which have to be done to manage the evaluation. Within this work, an automatic tool was implemented, which performs this action among others. Once the expected landmarks are available, it is possible to compare them systematical/y with the found landmarks of the landmark-detector. The comparison is made by different points of view. So, on the one hand, different error categories are determined. On the other hand, the deviations in time between expected and correctly found landmarks are also part of the evaluation. Finally, parameters of the landmark-detector were changed within this work to get possible additional indications of improvements. The work "Landmark- detection for distinctive feature-based speech recognition" by Sharlene A. Liu (1996) played a big role for the development of the given landmark­detector. So, this work also orients on Liu and the evaluation she made on her implementation.