The purpose of a ship lire alarm system is to detect a fire at an earlier stage in the ship and then to give a high reliable judgement result. This paper presents a multi-sensor fire detection algorithm using two fire parameters (temperature and smoke density). The processing of the sensor signals are carried out by fuzzy inference system based on neural networkgenetic algorithm learning method. The use of more than one sensor in the fire alarm system provides additional information on the environmental condition and allows a less erroneous alarm between the fire and no-fire case. The simulation results demonstrate the effectiveness of this method.