The invention discloses a multivariable time-lag parameter estimation method based on FIR (Finite Impulse Response) model identification, comprising the following steps of: step 101, collecting the input and output data of a system; step 102, carrying out the FIR model identification on the collected data to obtain a system step response coefficient and a step response curve; step 103, selecting threshold to be delta%(infinity); step 104, judging whether the fluctuation of a time-lag phase in the step response curve is in the threshold range or not; if the fluctuation does not exceed the threshold range, carrying out the step 105; otherwise, carrying out the step 103 to reselect a threshold; step 105, obtaining the first intersected point between the threshold and the step response curve; step 106, if t is judged to be larger than t1 and smaller than t2, carrying out a step 107 to obtain corresponding time-lag estimation value; otherwise, if the horizontal coordinator value t indicated by the intersected point is arranged on a sampling point t1 or t2, judging that the time-lag parameter estimation value zeta to be the sampling point t1 or t2; and step 107, obtaining the time-lag parameter estimation value zeta=t1. The invention has good applicability and high precision.