In forensic phonetics, one parameter commonly used in forensic voice comparisons is filled pauses (hesitant sounds). The aim of this study is to evaluate their discriminatory potential and, consequently, their usefulness in forensic voice comparisons so as to obtain reliable and robust voice identifications. In order to carry out this assessment, several variables related to the voice quality and the formant structure of these pauses are analyzed. The results show that the variables linked to the voice quality are more discriminatory than the variables related to the formant structure. Furthermore, the joint consideration of all the variables has achieved even better results than when those variables are considered separately. When a closed type known corpus is used, an average of 90% of correct classifications are obtained.