Methodologies of speech research with respect to the production processes are discussed, with an emphasis on the recent development of new instrumental techniques. It is argued that systematic studies of large amounts of speech data are necessary to understand the basic characteristics of speech. The traditional notion of phoneme-size segments seems inappropriate for interpreting multidimensional articulatory movements by a concatenative model. Experimental means such as a computer-controlled X-ray microbeam technique and advanced statistical processing, in combination with a new theoretical framework of phonetic description, promise future development.