An iterative grid message-passing algorithm for model-based digital image halftoning is introduced. Based on the standard message-passing algorithm on the grid graphical model, the algorithm is designed to suboptimally solve general two-dimensional (2D) digital least metric (DLM) problems and is found to be very successful (i.e., nearly optimal) for 2D data detection in page-oriented optical-memory (POM) systems. In contrast to many 2D (iterative) optimization techniques, this grid algorithm attempts to achieve a globally optimal solution via a local-metric computation and message-passing scheme. Using a reduced-complexity technique, the simplified grid algorithm is proposed for the halftoning problem and is shown to provide similar image quality as compared to the best halftoning algorithms in the literature. Since the grid algorithm does not exploit the properties of a specific metric, it is directly applicable to other digital image processing tasks (e.g., optimal near-lossless coding, entropy-constrained halftoning, or image/video dependent quantization).