In this paper is proposed a methodology for quantification of two veterinary hormones (17β-estradiol and noradrenaline) in dog serum based on surface-enhanced Raman spectroscopy (SERS) in conjunction with Random Forest for data treatment. It does not require sample pre-treatment and was automated using a multiwell plate holder from Raman equipment of 96 well plates, allowing the sequential analysis without manipulations in less than one hour. Comparable results with ELISA reference method were achieved, with relative errors of prediction not superior to 15% for 17β-estradiol and 7% for noradrenaline. Studies concerning calculations of systematic errors, tendency in the residuals, interval confidence and variable importance were performed permitting certification of the methodology validity and identification of SERS bands related to the hormones quantification.