The Liljencrants-Fant (LF) model is used to capture the shape parameters from the voice source. In this paper, two LF-model fitting approaches (one time-domain, one frequency-domain) are presented and compared by applying each to artificial and real speech source signals. Experimental results demonstrate that in most cases the time-domain method is superior to the frequency-domain based algorithm. By assessing approaches for estimating the LF-model parameters from a glottal source signal, this paper makes a contribution to the investigation of voice source parametrisation.