Acoustic analysis as used in the vocal pathology literature has come to mean any spectrum or waveform measurement taken from the digitised speech signal. The purpose of the work as set out in the present thesis is to investigate the currently available acoustic measures, to test their validity and to introduce new measures. More specifically, pitch extraction techniques and perturbation measures have been tested, several harmonic to noise ratio techniques have been implemented and thoroughly investigated (three of which are new) and cepstral and other spectral measures have been examined. Also, ratios relevant to voice source characteristics and perceptual correlation have been considered in addition to the tradition harmonic to noise ratios. A study of these approaches has revealed that many measurement problems arise and that the separation of the indices into independent measures is not a simple issue. The most commonly used acoustic measures for diagnosis o f vocal pathology are jitter, shimmer and the harmonic to noise ratio. However, several researchers have shown that these measures are not independent and therefore may give ambiguous information. For example, the addition of random noise causes increased jitter measurements and the introduction of jitter causes a reduced harmonic to noise ratio. Recent studies have shown that the glottal waveform and hence vibratory pattern of the vocal folds may be estimated in terms of spectral measurements. However, in order to provide spectral characterisation of the vibratory pattern in pathological voice types the effects of jitter and shimmer on the speech spectrum must firstly be removed. These issues are thoroughly addressed in this thesis. The foundation has been laid for future studies that will investigate the vibratory pattern of the vocal folds based on spectral evaluation of tape recorded data. All analysis techniques are tested by initially running them on specially designed synthesis data files and on a group of 13 patients with varying pathologies and a group of twelve normals. Finally, the possibility of using digital spectrograms for speaker identification purposes has been addressed.