AbstractAudio-tactile lane-marking (ATLM) is designed to alert inattentive drivers when they deviate from their lane. Most commonly used previously to mark the edges of divided freeways, it is increasingly used to mark the edge and/or centre of two-lane undivided roads. We present a detailed review of ATLM evaluations to demonstrate that although both edge- and centreline ATLM appear to be effective, evaluations have been flawed insofar as they have (1) often not controlled adequately for variations due to extraneous factors known to affect crash incidence; (2) often focussed on total crashes, or on inattention-related crashes, and failed to address the possibility that ATLM increases some crash-types (particularly crashes involving two-wheeled vehicles, out-of-control crashes, or crashes on the opposite side of the lane to which the ATLM is installed); (3) almost always related to the milled-in “rumble strips” employed in the US rather than the raised profile lane-marking (PLM) that is employed in Australia. The present Empirical Bayes analysis aimed to evaluate PLM by comparing treated and untreated sections of the same roads both before and after the date of treatment in terms a range of crash types (some separately for roads with only edgeline-PLM, with only centreline-PLM, and with both edgeline- and centreline-PLM). Although some of the above issues could not be addressed due to insufficient data, the analysis provides a useful template for future researchers. The results provide evidence that both edgeline and centreline (alone or in combination) reduce targeted (same-side) crashes. Although there is no evidence that edgeline and centreline alone increase opposite side crashes, only edgeline and centreline together produced a net reduction in total crashes. Thus, the present results suggest that installing PLM on both edgeline and centreline is preferable to installing either alone.