Flashover is extremely hazardous to both lives and properties.Different mathematical models in the form of differential equations have been developed for predicting the occurrence of flashover.The accuracy of the prediction is highly dependent on the sophistication of the mathematical models and the corresponding computational codes.Expensive computer resources are usually required for execution of the simulation programs.Over the past decade,the Artificial Neural Network (ANN)has been proven to be an efficient tool for the analysis of non-linear systems.Its applications cover a wide variety of fields for the prediction of physical phenomena and system behaviour,classification and decision making.In fire dynamics studies,a typical multi-layer perception in form of feedforward and recurrent network architectures has been applied successfully.In the current study,the Fuzzy-ARTMAP(FAM),which was developed by Grossberg in 1992,is used to determinethe occurrence of flashover for enclosures with different geometrical parameters and fire sizes.The FAM is a specially designed self-organizing ANN particular for supervised clustering.Data sets can be clustered into different categories by comparing the fuzzy subset-hood between input and stored patterns.this paper presents the FAM technique which has been employed to map the input fire parameters(i.e.input patterns) to a binary output (i.e.flashover or not).The computations are very cost-effectively in that the computational resources requored are much less as compared to other modelling techniques.Also,an extremely highaccuracy of prediction of the flashover in enclosures,a very non-linear system,has been achieved.