The aim of this work is to propose an innovative architecture for fire detection, using a wireless sensor network. This scheme is based on the statistical modeling of the Angstrom index. So far, such an index has been applied in outdoor environments. The results obtained by our analysis have extended its validity also to indoor scenarios. A key strength of our system is that each sensor node models the collected measures of the index, in order to transmit, in normal conditions, only the parameters of such a model, thus saving energy. To test the effectiveness of our fire detection system, we have simulated a fire in a building using the fire dynamics simulator. Results have shown that our proposed system is able to reach a good tradeoff between energy consumption and detection delay, in several operative conditions.