The mathematical model of harmonic state estimation and total least squares method are introduced. Harmonic state estimation is analyzed using robust total least squares method, considering the effect of measurement error and parameter error. This paper analyzes the shortcoming of total least squares that it is less capable of resisting gross error and proposes that the robust total least squares is used for harmonic state estimation. A mathematical model of IEEE-14 bus system is built in Matlab. Normal distribution error of gross error is added to measurement data, so as to normal distribution error to parameter matrix. The probability density curve and a comparison among total least squares, robust least squares and robust total least squares show that harmonic state estimation based on robust total least squares is the most accurate.