This paper proposed an object-based multiscale method for synthetic aperture radar (SAR) image change detection based on the statistical model. Rather than the pixel-based analysis conducted in the traditional way, the object-based image analysis was employed to take a collection of pixels as the unit of analysis, which reduced small spurious changes and was less strict relative to registration. In addition, a multiscale concept was adopted to exhibit the inherent multiscale characteristics of the target. To achieve object-based, multiscale change detection results, the multidate segmentation was performed on two temporal SAR images and extended to a set of suitable scales. Then, the Edgeworth series expansion was employed to estimate the probability density function, and the Kullback–Leibler divergence was adopted to calculate the distance between pairs of pixel collections. Next, the divergence index maps were divided into changed and unchanged classes to obtain change detection results for each scale. Finally, the subresults were combined to obtain a more accurate detection result. The experimental results obtained using real data demonstrated the effectiveness of the proposed method.