This paper describes an experimental study which investigates the viability of the currently accepted 'transcript method' in the preparation of voice (ear witness) line-ups. Two such line-ups were built by the Ottawa Police in a recent investigation and both these line-ups are experimentally examined for bias. Subjects (n=72, n=90) were asked to respond to a number of questions about the voices in the line-ups and in both cases the suspect's voice was judged significantly different from the foil voices. There appear to be a number of problems with this method- indeed it is not clear that it is capable of producing an unbiased line-up. Given our current knowledge about perceptual voice identification there appear to be good reasons for limiting the Canadian legal system's use of ear witness line-ups.