Most ATR methods deal with the single target. Actually, there are some targets arranged by some rules in spatial domain, which can be called "targets′ arrays". To recognize them, the spatial information among targets is used to make up the uncertainty of targets′ own features. A novel recognition approach of blob targets′ array at low resolution context (remote sensing images, infrared images, multi-spectrum images etc.) is presented. The method firstly utilizes the ARG model to customize the targets′ arrays and combine the spatial relationships of array nodes and the targets′ other simple attributes. By estimating the probability density distribution of the spatial relationships among targets and their own features, the arrays can be recognized. The method is different from the classic ATR method which is based on ARG model matching. Instead, the method extracts the probability features of spatial information from ARG model directly to recognize. The experiments show that the method is effective.