The adverse effects of benzodiazepines on driving are widely recognised. The aims of this study were both to determine the impact of naturalistic conversation on the driving ability of drivers under a benzodiazepine, and to measure the accuracy of drivers’ assessments of the joint effects of the benzodiazepine and conversation. Sixteen healthy male participants (29.69 ± 3.30 years) underwent a randomised, crossover, double-blind, placebo-controlled study with the benzodiazepine lorazepam (2 mg). They drove 200 km (125 miles) on a motorway in the morning. We measured two driving ability-related variables (i.e., lane-keeping performance), and collected a set of self-assessed variables (i.e., self-assessment of driving performance) during two 10-min sequences of interest (no conversation vs. conversation). An analysis of variance revealed an interaction whereby lane-keeping performance under lorazepam was worse in the no-conversation condition than in the conversation condition. No such difference was detected under placebo. Pearson's correlation coefficients revealed that self-assessments were (i) not at all predictive of lane-keeping when performed before the drive, but (ii) moderately predictive of lane-keeping performance when performed during or after the drive. We conclude that conversation with a passenger may contribute to safer lane-keeping when driving under a benzodiazepine. Moreover, a degree of awareness may be attained after some experience of driving under the influence of this type of medication.