An approach for a program is presented that automatically detects the position of pitch accents and intonation phrase (IP) boundaries from waveforms. The technique maps acoustic correlate bundles to prosodic labels by inspecting the course of F0 and intensity. Within an analysis frame specific criteria are applied to select possible candidates of pitch accents and IP-boundaries. The system works without any pre-segmentation of the speech waveform and does not use HMM-techniques. The program goes linearly through the F0-track and applies relative comparisons only within a time-limited analysis frame. Application of the program for intonational labeling of a large acoustic speech database and its possible use in automatic speech recognition systems is discussed.