The invention discloses a variational segmentation method based on multiple pieces of statistical information. The method includes the following steps: S1, aligning target binary images to a benchmark binary image; S2, acquiring a shape matrix M = {[phi]1, [phi]2, ..., [phi]n} of the aligned images, and carrying out principal component analysis to obtain an analysis shape pattern variation quantity; S3, utilizing a principal component analysis result to carry out shape expression on manually segmented binary images; S4, calculating an average cumulative distribution function on the manually segmented binary images; S5, selecting the adjacent manually segmented binary images, and calculating local statistical information fitting energy and global statistical information fitting energy; S6, segmenting the images according to data obtained in the S4 and the S5; and S7, executing the steps S3-S6 until all the manually segmented binary images are completely selected. According to the method, a target and a background in the images are effectively separated, the segmented images are obtained, phenomena of image under-segmentation, over-segmentation, overflow and the like are reduced, the computational redundancy is low, and the quality of image segmentation is guaranteed.