In horses, serum cortisol concentration is considered by searchers as an indirect measurement of stress. However, it includes the free and the transcortin-bounded fractions, and the sampling method is invasive. This is not the case for saliva cortisol, which represents a part of the free cortisol fraction, the biologically active form. The aim of this study was to compare salivary and serum cortisol assays in horses, on a large scale of concentrations using an ACTH stimulation test. In five horses, blood samples were drawn using intravenous catheter. Saliva sampling was made using swab. In saliva, cortisol was assayed by a direct radioimmunoassay (RIA). In serum, dilutions were necessary. Minimum detection limits were 0.2 nM and 8 nM respectively in saliva and serum. Precision and reproducibility of both RIAs were acceptable (CV<10%). Variables normality was assessed using the Shapiro-Wilk test. Transformation for variable normalisation was applied when necessary. Methods for pooling multiple subject data for linear analysis were applied (Chin-Sang Poon). At rest, cortisol concentrations were 188.81 ± 51.46 nM (mean ± sd) in serum and 1.19 ± 0.54 nM in saliva. Peaks were reached after 96 ± 16.7 min in serum (356.98 ± 55.29 nM) and after 124 ± 8.9 min in saliva (21.79 ± 7.74 nM) (Student t test, p<0.05). Percents of discharge were also different (225% in serum and 2150% in saliva, Student t test, p<0.05). The Pearson correlation coefficient between serum and saliva cortisol concentrations was 0.90 (adjusted R-squared=0.80) (p<0.001). The strong link existing between serum and saliva cortisol levels has been estimated by a regression analysis: Fserum = 159 + 56.7 loge (Fsaliva). Reliability of both RIAs and regression founded between serum and salivary cortisol levels permit the validation of saliva sampling as a non invasive technique for cortisol level assessment in horses.