With the recognition of herbal medicines, reliable and convenient methods for herbal medicines discrimination are needed. This paper introduces a novel method of using an electronic nose with online conformal prediction to classify 12 different types of herbal medicines with similar appearance. The performances of different online conformal predictors based on different training set updating strategies and varied sizes of initial training sets are evaluated to investigate the effectiveness of online conformal prediction. The results show that online conformal prediction manages to classify these medicines and achieves improved accuracy and robustness with more observations if the reliability requirement for training set updating is strict enough. Furthermore, the validity of online conformal prediction is vindicated that with the accumulation of observations, the error rate of prediction gradually converges below the significance level set by users, which offers users a flexible control over reliability and information about potential risk. Finally, the efficiency of online conformal prediction is discussed that customers should make a trade-off between reliability and efficiency.