The invention discloses a least square fitting dynamic frequency measurement method. The method comprises steps that A/D conversion sampling of power grid continuous signals is carried out to form equal time interval discrete signals; an FIR digital filter is employed to extract relatively pure fundamental wave discrete signals, and a sampling matrix is formed; a binary function Taylor expansion is utilized, frequency deviation and frequency change deviation of the power grid signals are extracted, and a constant coefficient matrix is calculated in an offline mode; the least square method is utilized to solve an unknown parameter matrix containing frequency deviation and frequency change deviation; dynamic frequency of each sampling point in the data window is solved. The method is advantaged in that based on the multivariate Taylor series, a power grid signal measurement matrix equation is established, the least square method is utilized to solve an equation linearity fitting problem, on the condition that frequency tracking is realized, a frequency change rate can be further monitored, and high accuracy, good timeliness and strong interference immunity are realized.