Carfentanil is a highly potent opioid legally used to tranquillise large animals (Wildnil®), but illicitly used as a recreational drug. In the 2002 Moscow Dubrovka theater hostage crisis, carfentanil was used as a knockout agent by Russian security forces, demonstrating its potential for use as an antagonistic chemical agent. Illegal use of carfentanil is considered a serious crime which is why all possible intelligence should be gathered for forensic investigation. Hence, dispersed carfentanil could be sampled from a crime scene during investigation. In a previous study, we showed that fentanyl agents can be sampled from surfaces and the associated synthesis method classified based on impurity profiles by using a multivariate classification model. In this study, we show that this classification method can also be applied for the highly-toxic carfentanil. Carfentanil was synthesised using three different methods and their chemical profiles analysed by GC-MS and UHPLC-HRMS. A training set of 54 samples was used to build a classification model of carfentanil synthesis methods based on impurity profiles. Synthesis methods for an independent test set were predicted by the model with good results. Furthermore, the model’s ability to classify by synthesis method was investigated using surface samples of carfentanil dispersed on cotton fabric and a wood surface collected after 72 h and one week. The results showed successful classification by synthesis route from both surfaces at both time points, with cotton fabric producing the best outcome.