Granular data classification and clustering is an upcoming and important issue in the field of pattern recognition. The paper proposes a granular neural network called as "Reflex fuzzy min-max neural network" for classification. Reflex mechanism inspired from human brain is exploited here to handle class overlaps. This network can be trained on-line using granular or point data. The proposed neuron activation functions are designed to tackle data of different granularity (size). Experimental results on real datasets show that the proposed algorithm can classify granules of different granularity more correctly compared to General Fuzzy Min max Neural network proposed by Gabrycz and Bargiela.