Mature microRNAs are short, single-stranded RNAs that bind to target mRNAs and induce translational repression and gene silencing. Many microRNAs discovered in animals have been implicated in diseases and have recently been pursued as therapeutic targets. However, conventional pharmacological screening for candidate small-molecule drugs can be time-consuming and labor-intensive. Therefore, developing a computational program to assist mature microRNA-targeted drug discovery in silico is desirable. Our previous work (https://doi.org/10.1002/advs.201903451) revealed that the unique functional loops formed during Argonaute-mediated microRNA-mRNA interactions have stable structural characteristics and may serve as potential targets for small molecule drug discovery. Developing drugs specifically targeting disease-related mature microRNAs and their target mRNAs would avoid affecting unrelated ones. Here, we presented SMTRI, a convolutional neural network-based approach for efficiently predicting small molecules that target RNA secondary structural motifs formed by interactions between microRNAs and their target mRNAs. Measured on three additional testing sets, SMTRI outperformed state-of-the-art algorithms by 12.9-30.3% in AUC and 2.0-18.4% in Accuracy. Moreover, four case studies on the published experimentally validated RNA-targeted small molecules also revealed the reliability of SMTRI. Our method is available as a user-friendly web server at http://smtri.net.