Commercially available computed tomography (CT) technologies such as iterative reconstruction (IR) have the potential to enable reduced patient doses while maintaining diagnostic image quality. However, systematically determining safe dose reduction levels for IR algorithms is a challenging task due to their nonlinear nature. Most attempts to evaluate IR algorithms rely on measurements made in uniform phantoms. Such measurements may overstate the dose reduction potential of IR because they don’t account for the complex relationship between anatomical variability and image quality. The purpose of this study was to design anatomically informed textured phantoms for CT performance evaluation. Two phantoms were designed to represent lung and soft-tissue textures. The lung phantom includes intricate vessel-like structures along with embedded nodules (spherical, lobulated, and spiculated). The soft tissue phantom was designed based on a three-dimensional clustered lumpy background with included low-contrast lesions (spherical and anthropomorphic). The phantoms were built using rapid prototyping (3D printing) technology and imaged on a modern multi-slice clinical CT scanner to assess the noise performance of a commercial IR algorithm in the context of uniform and textured backgrounds. Fifty repeated acquisitions were acquired for each background type and noise was assessed by measuring pixel standard deviation, across the ensemble of repeated acquisitions. For pixels in uniform areas, the IR algorithm reduced noise magnitude (STD) by 60% (compared to FBP). However, for edge pixels, the noise magnitude in the IR images ranged from 20% higher to 40% lower compared to FBP. In all FBP images and in IR images of the uniform phantom, noise appeared to be globally non-stationary (i.e., spatially dependent) but locally stationary (within a reasonably small region of interest). In the IR images of the textured phantoms, the noise was globally and locally non-stationary.