Synthetic aperture radar (SAR) chip segmentation is a crucial step in SAR automatic target recognition. When interested objects are placed on the ground, shadow is observed, and then the goal of SAR chip segmentation is to delineate target and shadow regions from background clutter. Due to fluctuations in SAR images, purely intensity-based methods lost many meaningful target and shadow regions. This drawback maybe conquered by introducing extra contextual information. The spatial relation between target and shadow regions is a kind of important contextual information, but rarely exploited in previous methods. In this paper, according to SAR imaging geometry, we conclude that target and shadow regions should be connected along the range direction. Based on this inter-connectivity between target and shadow regions, a new SAR chip segmentation method called SRPF-MRF is proposed. The new method introduces a Spatial Relational Potential Function (SRPF) term as constraint on the inter-connectivity between target and shadow regions, into Markov Random Field (MRF) based segmentation method. By SRPF-MRF segmentation, the segmented target and shadow is more complete, and therefore brings benefits to the subsequent feature extraction and object recognition. Finally experimental results on MSTAR dataset are given to show the superiority of SRPF-MRF method.