A method and system is provided for detecting occurrences of semantic temporal events based on observations extracted from input data and event models. The input data is fed into the system from some data source. Based on specified event to be detected, multiple-layer models corresponding to the event are retrieved. The models are used to determine the types of temporal observations to be extracted from the input data. The extracted temporal observations are then used, in combination with the multiple-layer models of the event, to detect the occurrences of the event.