Navigation plays a essential role in everyday life, and Virtual Environments (VEs) can serve as valuable tools for enhancing spatial learning. Extended Reality (XR), offering high-resolution immersive experiences, provides researchers with the ability to create controlled VEs for spatial learning. In this context, this thesis investigates the user's spatial learning by comparing photo-realistic and 3D modeled VEs with varying levels of realism during navigational tasks in XR. Employing an empirical research design, the study involved 12 participants solving navigational tasks in both types of VEs, with their spatial learning assessed through sketch maps and their sense of presence, perceived task difficulty, and overall preference assessed through questionnaires. Quantitative data analysis was employed, using Analysis of Covariance (ANCOVA) to adjust for potential covariates and the Wilcoxon Signed-Rank Test to compare differences in spatial learning outcomes between the two types of VEs. The findings suggest that while overall spatial learning was not significantly influenced by environmental realism, the recall of complex object associations was markedly enhanced in more photo-realistic settings. This enhancement underscores the potential benefits of high-fidelity graphics in designing XR applications for educational and training purposes, particularly where complex spatial details are crucial. Participants reported a greater sense of presence in more realistic environments, although this did not translate to differences in perceived task difficulty or overall preference. The study further contributes to understanding the role of realism in VEs, offering insights for developers and educators in optimizing XR environments for enhanced user engagement and learning efficacy.