Digital halftoning is required whenever a grayscale image must be rendered on a device with fewer graylevels, such as a fax machine, laser printer, or low cost liquid crystal display. Error diffusion is a halftoning method which produces excellent results. However, its operation is not fully understood, and methods for improving its performance and evaluating the results are scarce. In this talk, we show how error diffusion is equivalent to the noise shaping feedback coder, a form of delta-sigma modulator. We use this correspondence to explain the sharpening seen in error diffused imagery. We further devise two image quality metrics, one for quantization noise and one for sharpening, that accurately quantify the performance of a particular error diffusion kernel.