Dual-energy computed tomography (DECT) has been widely used in many applications that need material decomposition. Image-domain methods directly decompose material images from high- and low-energy attenuation images, and thus, are susceptible to noise and artifacts on attenuation images. To obtain high-quality material images, various data-driven methods have been proposed. Iterative neural network (INN) methods combine regression NNs and model-based image reconstruction algorithm. INNs reduced the generalization error of (noniterative) deep regression NNs, and achieved high-quality reconstruction in diverse medical imaging applications. BCD-Net is a recent INN architecture that incorporates imaging refining NNs into the block coordinate descent (BCD) model-based image reconstruction algorithm. We propose a new INN architecture, distinct cross-material BCD-Net, for DECT material decomposition. The proposed INN architecture uses distinct cross-material convolutional neural network (CNN) in image refining modules, and uses image decomposition physics in image reconstruction modules. The distinct cross-material CNN refiners incorporate distinct encoding-decoding filters and cross-material model that captures correlations between different materials. We interpret the distinct cross-material CNN refiner with patch perspective. Numerical experiments with extended cardiactorso (XCAT) phantom and clinical data show that proposed distinct cross-material BCD-Net significantly improves the image quality over several image-domain material decomposition methods, including a conventional model-based image decomposition (MBID) method using an edge-preserving regularizer, a state-of-the-art MBID method using pre-learned material-wise sparsifying transforms, and a noniterative deep CNN denoiser.