The fuzzy inference system (FIS) has been developed for predicting customer buying behavior. Three different methods: (grid partitioning, fuzzy c-means, subtractive) have been used to get the membership values during the fuzzification of inputs which is the first step in the creation of FIS. For each method, two different FIS models (Mamdani-type FIS and Sugeno-type FIS) have been developed. ANFIS training is also done on the Sugeno-type FIS to tune the FIS parameters using the input/output training data. Finally, the comparison table has been prepared to list out the efficiencies in terms of accuracy for the different techniques used and thus finds out which method is the best for the particular system.