A technique for detecting spills is described. A number of audio sensors detect a sound potentially associated with an object spilling on a surface and the audio data from the sensors is analyzed in order to generate an acoustical analysis of the sound. The analysis of the sound detected by the audio sensors is compared against a database of known sounds corresponding to a number of types of spill incidents. Based on a match detected between the analysis of the sound and a type of spill incident in the database of known sounds, the identity of the object spilled can be determined. The location of the spilled object can also be calculated by triangulating the sensor data from the audio sensors.