The following problems that results from logarithmic transformation linearizes power or exponential model are generally ignored:(1) Which model structures are linearized with logarithmic transformation? (2) Could the error properties be changed under the transformation? (3) Would the regression coefficients of estimate with the transformation be unbiased and minimum-error-variance? Such problems and appropriate improvements are discussed in this paper.