Abstract: The operating temperatures in Molten Carbonate Fuel Cells (MCFC) stack must be controlled within a specific range. The efficiency of the stack will be lower and the lifespan of the stack will be shorter if operating at too high or too low temperature. So it is very important to control the temperatures in the stack. Unfortunately, for the seriously complexes of MCFC, modeling MCFC is very difficult. The models existed are all the analytical models described with complicated nonlinear differential equations and poor for online-control. So, In this paper, firstly, a neural networks identification model of MCFC stack is developed based on measured input-output sampled data. Then, an online novel adaptive fuzzy controller of MCFC stack is developed. Some drawbacks of traditional fuzzy control method are improved. On the one hand, the parameters of the fuzzy system are regulated, adopting BP algorithm, on the other hand, the rule database of fuzzy system is adaptively adjusted, adopting the nearest neighbor-clustering algorithm. In the end, using the neural networks model of MCFC stack as the real MCFC stack the control simulations is carried out. The validity of neural networks identification modeling of MCFC stack and the superior performance of the novel adaptive fuzzy controller are demonstrated by simulation.