Today's cyber-physical systems (CPS) like advanced driver assistance systems (ADAS) in modern vehicles use a large variety of sensors to process data from their surroundings. Thereby, systems like lane departure warning and adaptive cruise control support the driver in tedious or critical traffic situations. During the development of such systems, engineers also use recorded sensor data in offline validations to complement simulations that exhibit optimal environmental conditions. Such recordings are an ever-growing data source, and thus, effective methods are needed to find proper recordings in databases to support the system validation. Textual annotations for these sensor recordings require a well-defined taxonomy and continuous maintenance. Instead of relying on such a manually maintained taxonomy, an automated method for identifying relevant scenarios from real world sensor recordings by using simulation models is described. The outlined approach is evaluated with real world data sets used by lane-detection algorithms from nine different projects. Results from these data sets of more than 2.3 GB show that finding relevant traffic scenarios is possible in less than 0.15s.