A new inverse filtering technique for the estimation of the glottal excitation is presented in this study. The method is based on the application of a new algorithm, called Discrete All-pole Modeling (DAP), used in the estimation of the vocal tract transfer function. The paper presents results that were obtained from natural and synthetic vowels of different phonation types and fundamental frequencies. It was found that estimated glottal waveforms comprised less formant ripple when the inverse filtering analysis was performed with the DAP-technique instead of conventional linear prediction. Glottal waveforms computed from natural female voices were also characterized by longer glottal closed phases when the DAP-technique was used.