Web pages nowadays constitute the most popular source of information, business and entertainment provision. Inarguably, their aesthetics comprise an integral part of the design of a website, playing a multidimensional role. Initially, web aesthetics support the content and functionality of a website, while at the same time striving to pique the interest of the targeted user categories. The objectives of this diploma thesis are to investigate and highlight the importance of demographic characteristics when evaluating web design aesthetics, through the use of deep learning algorithms. For this, two different approaches have been applied. The first approach concerns the training of three different architectures of convolutional neural networks (CNN) across the available data set, the AlexNet, VGG16 and Xception architectures. AlexNet has been re-evaluated on this set and provides reliable results while VGG16 is presented as an improved solution. On the other hand, Xception is a contemporary architecture which is being tested for the first time on this dataset and has surpassed the literature results. The second approach involves splitting the dataset by demographic groups and training convolutional networks for each group separately. In this way the respective models can model the aesthetics preferences of each demographic group. These models are merged using various ensemble methods and the best one is opted for the evaluation and comparison of the findings. In the experiments performed, comparisons are made between the models of each approach, as well as a presentation of various relative examples is given for better understanding. The purpose of this thesis is to point out the determinant role and importance of demographic characteristics, while also highlighting the contribution of advanced deep learning algorithms to the achievement of reliable predictive results regarding subjective issues, such as web site aesthetics.