Abstract: In this paper a fast learning algorithm is presented for fuzzy logic system. The number and the distribution of fuzzy sets in input variables must be previously settled. The antecedent part of fuzzy rules could be any form of fuzzy sets and the consequent must be singleton form; Fuzzy inference adopts product method. Defuzzification is Tsukamoto method. Fuzzy rules are obtained from the input-ouput data pairs. The value of fuzzy rule' consequent is calculated just one time by least square method. The precision that it could reach depends on the the number of fuzzy sets in input variables. This algorithms consumes realtively less computing power than other learning algorithms.