Proposed are a system (1) and a method for the determination and forecast of absolute and relative risks for car accidents based on exclusively non-insurance related measuring data and based on automated traffic pattern recognition, wherein data records of accident events are generated and location-dependent probability values for specific accident conditions associated with the risk of car accident are determined. Thus, the proposed system (1) provides a grid-based (2121, 2122, 2123, 2124), technically new way of automation of risk-prediction related to motor accidents using environment based factors (elevation, road network, traffic data, weather conditions) including socio-economic factors that are impacting motor traffic and are location dependent received from appropriate measuring devices and systems (41, . . . , 45). In this way, predictions of the accident risk for arbitrary areas can be provided. The system is calibrated by comparing features of areas or road segments with the number and type of accidents that have measured or registered there, linking the features and accident data e.g. using the below discussed machine learning techniques.