In recent years, the types and quantities of fentanyl analogs have increased rapidly. It has become a hotspot in the illicit drug control field of how to quickly identify novel fentanyl analogs and to shorten the blank regulatory period. At present, the identification methods of fentanyl analogs that have been developed mostly rely on reference materials to target fentanyl analogs or their metabolites with known chemical structures, but these methods face challenges when analyzing new compounds with unknown structures. In recent years, emerging machine learning technology can quickly and automatically extract valuable features from massive data, which provides inspiration for the non-targeted screening of fentanyl analogs. For example, the wide application of instruments like Raman spectroscopy, nuclear magnetic resonance spectroscopy, high resolution mass spectrometry, and other instruments can maximize the mining of the characteristic data related to fentanyl analogs in samples. Combining this data with an appropriate machine learning model, researchers may create a variety of high-performance non-targeted fentanyl identification methods. This paper reviews the recent research on the application of machine learning assisted non-targeted screening strategy for the identification of fentanyl analogs, and looks forward to the future development trend in this field.