This application discloses a facial image-based risk recognition method and apparatus, a computer device, and a storage medium, to resolve a problem of inaccurate recognition caused by occluding a face with a hand during video data analysis. The method includes: obtaining video data recorded when a customer answers a preset question; performing image recognition on the video data by using a pre-trained occlusion recognition model, where a recognition result includes hand-occluded facial images and hand-unoccluded facial images; obtaining a hand-occluded facial position corresponding to each hand-occluded facial image, and obtaining a tension degree recognition result based on hand-occluded facial positions corresponding to all the hand-occluded facial images; inputting the hand-unoccluded facial images into a pre-trained security determining model to obtain a risk degree recognition result; and determining a risk warning recognition result based on the tension degree recognition result and the risk degree recognition result, where the video data is risk warning data if the risk warning recognition result is greater than a first preset threshold.