In an image processing apparatus, fast error diffusion processing is performed without increasing the size of an error diffusion processing circuit, even in cases where the print data to be generated is 12 colors data or similarly large data. More specifically, with quantization by error diffusion, binarized image data for 12 colors is obtained by causing an image processing circuit that executes 6 colors quantization processing to operate twice. At this point, RAM realized by DRAM with comparatively slow operating speeds, for example, is accessed for error data diffused to a first raster and error data diffused from a last raster. In contrast, an error buffer realized by SRAM inside the image processing circuit with fast operating speeds is accessed for all other error data. Thus, as a result, memory access speeds for reading out and writing error data can be improved overall.