Curriculum Learning is an effective approach to training neural networks, consisting of a scoring function that determines and sorts the data complexity, and a pacing function that controls the data provided at each training epoch. Despite good results in different tasks, its applicability on Image-Denoising Autoencoders has not yet been fully validated. Therefore, in this paper, we investigate whether the Curriculum Learning strategy can be used as an improved alternative to conventional Image-Denoising Autoencoders training. We explored different pacing and scoring functions, comparing them with the conventional training method. Our results, performed on different benchmark datasets and cross-domain scenarios, were superior to traditional training, reducing processing time, and improving denoising.