Validating autonomous vehicle (AV) systems is a critical yet complextask due to the wide range of unpredictable and diverse situationsthat can occur in real-world driving. AVs must be capable of handlingnot only routine conditions, but also rare and dynamic interactionsthat involve other road users, environmental factors, and road infrastructure. To manage this complexity, researchers rely on scenarios,which are structured representations of driving situations that reflectthese diverse conditions. Scenarios play a central role in assessingAV performance by capturing key elements such as vehicle behavior, interactions with surrounding traffic, and the characteristics ofthe driving environment. Through systematic identification and analysis of scenarios, researchers can evaluate how AV systems respondto different challenges and improve their overall safety and reliability [19]. However, certain corner-case scenarios can be challenging todetect, as they are often rare and hidden within large-scale datasets.These scenarios are of particular interest because they represent situations where AV systems may fail or behave unpredictably, makingthe identification of these corner cases crucial for targeted testing andvalidation.This thesis proposes the following methods for detecting criticalscenarios in data collected from AV test vehicles: Interquartile Range,Isolation Forest with Dynamic Time Warping, and Sudden ChangeDetection algorithm. These methods were applied to a dataset thatcontains the dynamics of the ego vehicle, with the aim of detectingscenarios that represent corner cases. Their performance was studiedin two different road environments, i.e., urban and highway, providing insights into their respective strengths and weaknesses.In this thesis, we also extended these methods by integrating theminto a voting ensemble system, through which we achieved improvedaccuracy, coverage, and scalability in identifying critical scenarioscompared to the individual methods. The final ensemble methodachieved an accuracy of 90 % and 92 % in urban and highway environments, respectively, while remaining inexpensive and simple to implement. These unsupervised methods are minimally based on domainspecific knowledge and offer a more scalable and computationallyefficient alternative to manual scenario detection techniques.