The invention discloses a Bayesian small sample learning-based surface defect detection method. The method comprises the steps of firstly performing image collection on a detected object in industrialproduction by adopting a linear array industrial camera; converting a collected detected object image into a single-channel grayscale image from an RGB color space; performing sliding window filtering on the detected object image to remove noisy points in the detected object image; performing enhancement processing on defects in the detected object image by using a sobel operator; and selecting anormal training sample and a defective training sample subjected to enhancement as to-be-detected samples in the to-be-detected image, calculating out gradient features of the to-be-detected samplesand performing learning. The calculation processing can be performed according to the gradient features of the detected object by utilizing a Bayesian algorithm in combination with other steps, so that the detection performance and accuracy are improved, the labor can be effectively reduced, the labor intensity is reduced, the working efficiency is improved, and the detection precision is relatively high.