Vertebrate vocalizations are widespread secondary sexual signals used for mate attraction and territory defence, and variation in signal quality is often condition dependent and impacts reproductive outcomes. Although vocal signal performance is known to reflect various aspects of male quality, few studies have examined the underlying mechanisms mediating its costs and hence its honesty. Using a population of Arctic-breeding snow buntings (Plectrophenax nivalis), we compared the 'Oxidation Handicap Hypothesis', which predicts that testosterone-induced increases in oxidative stress provide a direct mechanistic basis for ensuring the honesty of many secondary sexual signals, to the 'Aerobic Activity Hypothesis, which predicts that it is the aerobic activity involved with signal production (i.e. vocal performance or defending a large territory) and not testosterone directly that links signal quality and oxidative stress. Males singing at faster rates had higher levels of both reactive oxygen metabolites and non-enzymatic antioxidant capacity in the plasma (i.e. without an increase in overall oxidative stress), enabling certain males to produce high-quality signals while also mitigating the costs of an associated increase in oxidative stress. However, these results were completely independent of plasma testosterone levels, supporting the role of aerobic performance in directly affecting oxidative stress. Although song performance was not linked to reproductive parameters in our data set, our research is the first to test these competing hypotheses in a behavioural trait and results suggest that oxidative stress may be an underlying physiological cost preventing low-quality individuals from producing high-quality signals.