Several types of duplex grain size distributions in five different alloys were evaluated using image analysis. Most of the grain structures contained annealing twins. Those with straight interfaces could be recognized and deleted from the image, leaving only grain boundaries. One specimen exhibited curved twin boundaries, caused by deformation, and they could not be discriminated by the system as currently programmed. Grain areas were measured and grouped according to their relationship to the ASTM grain size scale. An area-weighted histogram was shown to be excellent for revealing the nature of the distribution, while a numerical-frequency histogram was insensitive. The intersection of these two curves separated only one of the four bimodal distributions. A deconvolution approach, using the area-weighted curve only, should be evaluated. An arithmetic grain area classification approach using 25 classes based on the data range, to split the two grain area populations based upon the intersection of the number percent and area percent curves, worked well for two of the four specimens. Image analysis detection of grains results in a small portion of the image (about 6-12%) assigned to the grain boundaries. In manual measurement methods, the area occupied by the grain boundaries is not considered, and it does not influence measurements. Thus, compared to manual methods, image analysis undersizes grains slightly producing a relatively small positive bias in the grain size number, which could be ignored, but can be eliminated or reduced.