The objectives of acoustical analyses of speech are generally to obtain economical descriptions of the important characteristics of speech sounds, together with the distributions of these characteristics. The relevant acoustical properties cannot be determined and quantified easily without reference to data on the processes of speech production and perception. An understanding of speech generation indicates the constraints that are imposed on the vocal output, whereas knowledge of the perception process suggests acoustical properties that are likely to be important for understanding speech, as well as the accuracy with which they should be specified. It is probable that relatively inaccurate measurements of several “unreliable” features are adequate to distinguish reliably a given speech sound from others of the same language, provided these features are selected properly. These remarks are illustrated by existing data on the acoustical properties of vowel and consonant sounds, and areas are noted where further data are needed. The possibilities of utilizing the high-speed digital computers in the analysis of speech are discussed.