In this paper, a new segmentation method of segmenting unconstrained handwritten numeral string with unknowing number of digits is proposed, which mainly based on recognition-based segmentation, combining with dissection method and holistic method. We proposed an approach that gives out a simple, effective way to select candidate segmentation positions through finding out the start points and end points of ligatures. The approach is mainly based on the feature of variety of upper and lower contours, combined with the feature of vertical projection value. We also provide a method to confirm the correct segmenting points and to decide which digit the connecting stroke belongs to by using the dynamic programming. Aiming at decreasing the work-time and complexity of segmentation, we provide dissection method as well as the holistic method for segmenting some special sub-images. The experimental results will be given in the end.