We present an approach that detects physical events such as a fire or an explosion using sensor data fusion, where not all relevant signals describing the event are available due to non-presence or malfunctioning of some sensors. We employ abductive probabilistic reasoning to detect the occurrence of an event amongst several alternative events from imperfect sensor data. Influenced by Dempster-Shafer’s evidence theory, we reason on the available evidence produced by the sensor data, combined with counterevidence, to establish degrees of confidence to the different hypotheses made about the occurrence of an event. The paper also describes an experimental sensor setup for detection of fire and explosion events, and its effectiveness in terms of false negative and false positive detection rates.