Speech modeling is an active research area which has received considerable attention in recent years. Application areas include speech coding, recognition and synthesis. Several models have been proposed both in the time and frequency domain contexts. Early frequency domain models approximated the voiced segments by locally periodic signals. The generalized harmonic model was able to deal with departures from periodicity resulting from vocal tract variation, but had difficulty in handling fast pitch variations. Sinusoidal models were developed to overcome these limitations. In this paper, we briefly review the basic concepts related to sinusoidal modelling of speech and present the results of an extensive experimental comparison between several sinusoidal modeling variants.