The existing solar cell surface defect detection algorithms based on machine vision are all designed to use various types of mathematical models to carry out the algorithm design. In order to further improve the detection accuracy, inspired by human vision bionics, the human visual attention mechanism is firstly introduced in the solar cell surface defect detection, and a solar cell surface defect detection algorithm based on visual saliency is proposed in this paper. First of all, the acquired solar cell surface image is preprocessed to remove the noise and grids that influence the defect detection. Secondly, a visual saliency detection algorithm based on selflearning features is put forward to roughly locate the defect region. Then, an algorithm that combines the visual saliency and superpixel segmentation is proposed to precisely locate the defect region. At last, the final detection result is obtained using morphological postprocessing. The subjective and objective experiment evaluations on a test image database containing various types of defects demonstrate that the proposed algorithm has high detection accuracy.