A direct analysis in real time (DART) coupled with high-resolution mass spectrometry (HRMS) method was established to detect organic impurities in cocaine samples seized in China. This approach allowed discrimination between linked and unlinked seizures with quick sample preparation and high throughput. The responsivities of the mass spectra of 20 organic impurities found in 47 cocaine samples were explored as factors in multivariate analysis. Based on the data from 20 organic impurities in 47 linked/unlinked cocaine samples, 25 combinations of data pretreatment methods and distance/correlation measurements were compared to choose the combination with the best discriminating power for cocaine profiling. Data pre-treatment by calculating the logarithm (L) followed by Squared Euclidean distance (Squ Dij) was chosen as the best combination for the optimal discrimination of linked and unlinked samples. Hierarchical cluster analysis (HCA) was used to analyse another 46 cocaine samples, and 38 of the 46 cocaine samples were divided into 9 linked groups. Relevant information was provided for forensic science intelligence and case analysis. Display omitted •Twenty organic impurities were found in 47 cocaine samples seized in China by DART-HRMS.•Twenty-five combinations of measurements were tested to find the combination with the best discriminating power, and hierarchical cluster analysis was used.•DART-HRMS was used to analyse actual cocaine samples, and the correlation information of cocaine samples was provided to the police.