Statistical phonetic data for Polish were collected. Phonemes are of different lengths, varying from 30 ms to 200 ms. Average phoneme durations are presented. A corpus of spoken Polish was used to collect statistic values of real language and evaluated to be applied in an automatic speech recognition and speaker identification systems. These natural phenomena could be used in phonemes parametrisation and modelling. An additional source of information for a case of speech segmentation was obtained. The collected data are presented in the paper (average values for all available male speakers and for some chosen ones), along with comments on the corpus and the used method. The obtained data were compared with the expected values according to phonetic literature.