Recent research literature on the effectiveness (or otherwise) of promoting road safety via mass media advertising is selectively reviewed. The overall picture from this is inconclusive: effects of substantial size have been rare, but effects of small size cannot be ruled out. It is then argued that attempts to use crash data to establish or disprove the cost-effectiveness of campaigns are, indeed, doomed to failure: the random variability in crash numbers is too great (and even low effectiveness may be sufficient as campaigns are very cheap per person reached). It is suggested that evaluation may need to be based on before and after comparison of behaviours or variables that can be objectively observed and are closely linked to safety – and in order to substantiate the behaviour-safety link, credible theories are needed.