The traditional regression is based on the assumption that the uncertainty is from the randomness of the variables, but in fuzzy regression, the uncertainty is assumed from membership grades. Generally, fuzzy regression adopts the fuzzy variables of sample (X(subscript i), Y(subscript i)) to estimate the parameter of fuzzy regression while Y(subscript i) is the fuzzy variable and X(subscript i) is still the real number. We think that the assumption of X(subscript i) can not express factually all the information of the sample, so we assume X(subscript i) is the fuzzy variable in the study. In this way, the explanation of the sample will be pressed close to the reality and by using the least square approximation in estimation process, the original essence of regression is preserved but more investigation in fuzzy variables is given.