In a recent publication, Fox et al. (2019) described a three-year trial of a ‘virtual fence’ installed to reduce wildlife roadkills in north-eastern Tasmania. The authors reported a 50% reduction in total roadkills, concluding that the ‘virtual fence’ had the potential to substantially reduce roadkill rates. The field of roadkill mitigation has a long history of promising techniques that are ultimately found wanting, so we evaluated the conceptual basis of the ‘virtual fence’ and the design and analysis of the trial. Of the two stimuli emitted by the ‘virtual fence’, its lights only partly match the sensory capabilities of the target species, its sound frequency is suitable but the intensity is unknown, and both stimuli are artificial and lack biological significance, so will be prone to habituation once novelty wanes. The trial, conducted in three phases, revealed a total of eight methodological flaws ranging from imprecise measurements, confounding effects of treatments, low statistical power, violation of test assumptions and failure to consider habituation. Greater caution is needed in interpreting the findings of this study, and well designed, long-term trials are required to properly assess the ‘virtual fence’.