The Psychomotor Vigilance Test (PVT) is a widely used assay of behavioural alertness sensitive to the effects of sleep loss and circadian misalignment. However, there is currently no accepted PVT composite outcome metric that captures response slowing, attentional lapses and compensatory premature reactions observed typically in sleep-deprived subjects. We developed a novel likelihood ratio metric (LRM) based on relative frequency distributions in 50 categories of reaction times (RT) and false starts in alert and sleep-deprived subjects (acute total sleep deprivation: n = 31 subjects). The LRM had the largest effect size both in a 33-h total sleep deprivation protocol [1.96; 95% confidence interval (CI): 1.61–2.44; followed by response speed 1/RT, effect size 1.93, 95% CI: 1.55–2.65] and in a chronic partial sleep restriction protocol (1.22; 95% CI: 0.96–1.59; followed by response speed 1/RT, effect size 1.21, 95% CI: 0.94–1.59; 5 nights at 4 h sleep per night; n = 43 subjects). LRM scores correlated highly with response speed (R2 = 0.986), and less well with five other common PVT outcome metrics (R2 = 0.111–0.886). In conclusion, the new LRM is a sensitive PVT outcome metric with high statistical power that takes subtle sleep loss-related changes in the distribution of reaction times (including false starts) into account, is not prone to outliers, does not require baseline data and can be calculated and interpreted easily. Congruence between LRM and PVT response speed and their similar effect size rankings support the use of response speed as the primary, most sensitive and most parsimonious standard PVT outcome metric for determining neurobehavioural deficits from sleep loss.