AutoEncoders (AEs) are widely used for efficient data compression and feature extraction. However, finding the most appropriate architecture and set of parameters that result in the most effective approach for a certain domain is not an easy task. This study is inspired by the significant potential of using NeuroEvolution (NE)NeuroEvolution (NE) to optimize and develop AEs through evolutionary algorithms. For that purpose, a framework based on Fast Deep Evolutionary Network Structured Representation (Fast-DENSER) is introduced for automatically training and evolving an AutoEncoder (AE)AutoEncoders (AEs), aimed at offering Deep Neural Network (DNN)Deep Neural Networks (DNNs) architectures potentially adept at addressing a range of tasks such as data denoising, anomaly detection, and high-resolution image reconstructionImage reconstruction. The methodology involves intricate optimization and time-intensive exploration for identifying an optimal architecture tailored to a specific task, with the susceptibility to overfitting influencing result interpretation. Our approach involves training and evolving the encoder and decoder components of the AEAutoEncoders (AEs) separately. The experimental findings showcase diverse architectures, shedding light on the types of AEAutoEncoders (AEs) architectures favorable and unfavorable to effective training and evolution. The experiments also demonstrate the use of different fitness functions and the metrics that are most and least suited to image reconstructionImage reconstruction problems, along with their influence on the training and evolution process.AutoEncoders (AEs)NeuroEvolution (NE)