Today, the generation of true orthophoto mosaic with high quality and accuracy from UAV images which is a challenging problem has attracted the attention of many researchers. The selection of the best reference image is a crucial aspect that influences the accuracy and quality of this product. In UAV-based photogrammetry projects, there are several options among the overlapping images to construct each pixel of the orthophoto mosaic. Thus, the orthophoto mosaic result can be greatly affected by using various UAV images with various geometric and radiometric properties. This paper presents a new method that uses a Multi-Criteria Decision-Making method to identify the best reference image for each pixel of the orthophoto mosaic by considering multiple criteria. The optimal reference image is determined in two steps: weighing and Multi-Criteria Decision Making. In the weighting stage, the 3D tie points are used to determine the weight of five different criteria including the distance from the projection center, the accuracy of the external orientation parameters, the number of the 3D tie points, the number of the ground control points, and the image quality. Then, the images are ranked using a Multi-Mriteria Decision Making process once the closest images displaying the DSM point are found. The internal evaluation is then carried out to choose the best image by minimizing the reprojection error of the nearby 3D tie points. The final result of this process is an optimized reference image for each DSM point, which is used to fill the corresponding orthophoto mosaic cell. Two UAV-based datasets were used to analyze the proposed method, and the results are compared to those of the popular single-criteria approaches. The results show that, based on the results obtained, in terms of RMSE, the proposed MCDM method, reduces the errors generated on the orthophoto mosaic up to 15 cm, 10 cm and 8 cm compared to the single-criteria that employs the viewing angle, the distance from the projection center and the distance from Nadir. In the second dataset these amounts are 7 cm, 3 cm and 3 cm, respectively. Nevertheless, in the first dataset the processing time of the proposed method is about 28% longer than the fastest single-criteria method of distance from the imaging center and 15% longer in the second dataset.