BACKGROUND: The unprecedented proliferation of new psychoactive substances (NPS) threatens public health and challenges drug policy. Information on NPS pharmacology and toxicity is, in most cases, unavailable or very limited and, given the large number of new compounds released on the market each year, their timely evaluation by current standards is certainly challenging. AIMS: We present here a metabolomics-targeted approach to predict the pharmacological profile of NPS. METHODS: We have created a machine learning algorithm employing the quantification of monoamine neurotransmitters and steroid hormones in rats to predict the similarity of new drugs to classical ones of abuse (MDMA (3,4-methyl enedioxy methamphetamine), methamphetamine, cocaine, heroin and Δ9-tetrahydrocannabinol). RESULTS: We have characterized each classical drug of abuse and two examples of NPS (mephedrone and JWH-018) following alterations observed in the targeted metabolome profile (monoamine neurotransmitters and steroid hormones) in different brain areas, plasma and urine at 1 h and 4 h post drug/vehicle administration. As proof of concept, our model successfully predicted the pharmacological profile of a synthetic cannabinoid (JWH-018) as a cannabinoid-like drug and synthetic cathinone (mephedrone) as a MDMA-like psychostimulant. CONCLUSION: Our approach allows a fast NPS pharmacological classification which will benefit both drug risk evaluation policies and public health.