Cortisol is often used as a stress indicator in animal behaviour research. Cortisol is commonly measured in plasma and can also be measured in saliva. Saliva contains only the free form of cortisol, which is biologically active, and saliva sampling is not invasive and may therefore be less stressful. Our study aims to guide the choice between the measurements of cortisol in plasma v. saliva depending on experimental conditions. We analysed the effect of the level of cortisol in plasma on the concentration of cortisol in saliva compared to plasma and the effect of saliva sampling v. jugular venepuncture on the cortisol response. In Experiment 1, blood and saliva were collected simultaneously under conditions in which the expected cortisol release in blood varied: in an undisturbed situation or after the isolation of lambs from their pens or the administration of exogenous ACTH (six animals per treatment). In Experiment 2, we subjected lambs to saliva sampling, venepuncture or neither of these for 8 days to evaluate how stressful the sampling method was and whether the animals habituated to it by comparing the responses between the first and last days (four animals per treatment). All animals were equipped with jugular catheters to allow regular blood sampling without disturbance. Samples were collected 15 min before any treatment was applied, then at various time points up to 135 min in Experiment 1 and 45 min in Experiment 2. In Experiment 1, we observed a strong correlation between salivary and plasma cortisol concentrations (r = 0.81, P < 0.001). The ratio between salivary and plasma cortisol concentrations was 0.106 on average. This ratio was higher and more variable when the cortisol concentration in plasma was below 55 nmol/l. In Experiment 2, venepuncture induced a larger cortisol response than saliva sampling or no intervention on day 1 (P < 0.02); this difference was not observed on day 8, suggesting that sheep habituated to venepuncture. We recommend the measurement of cortisol in saliva to avoid stressing animals. However, when the expected concentration in plasma is below 55 nmol/l, the cortisol in saliva will reflect only the free fraction of the cortisol, which may be a limitation if the focus of the experiment is on total cortisol. In addition, if cortisol is measured in plasma and blood is collected by venepuncture, we recommend that sheep be habituated to venepuncture, at least to the handling required for a venepuncture.