The discrimination and identification of complex mixtures remain a significant challenge to chemical analysis. The conventional technique for complex mixture analysis refers to a complete component-by-component approach, such as gas chromatography/mass spectrometry (GC/MS), which requires sophisticated facilities and professional personnel. In this work, we propose a strategy using carbon nanotubes-based multichannel e-nose for complex mixture discrimination, taking coffee aroma as an example. By extracting efficient features from the sensing response profile, a highly distinctive smellprint feature for coffee aroma is achieved. In combination with an efficient machine learning classifier algorithm, an excellent identification accuracy of 97.4% for three types of coffee aroma is achieved. This proposed strategy provides a portable, lowcost, high-efficiency solution for complex mixture discrimination and could be applied in various fields, such as food quality monitoring, volatile organic compound-related disease diagnosis, environmental monitoring, public safety securing, etc.