The invention discloses an online statistical analysis technique for generator set primary frequency modulation performance abnormity. The method comprises the following steps: selecting typical primary frequency modulation events beneficial for improving precision through generator PMU measured data obtained online, calculating primary frequency modulation contribution rate of a corresponding set, calculating actually measured average speed change rate by adopting an integral algorithm, screening out reasonable primary frequency modulation performance evaluation results by using a reasonable data range and normal distribution variance, and shielding the influences of disturbance and secondary frequency modulation on primary frequency modulation performance evaluation; and then giving an alarm for primary frequency modulation abnormity of the set according to the difference between the two indexes and a reasonable value and a preset value and the change relative to historical value. Operating personnel can discover the generator sets with the problem of primary frequency modulation in time according to the calculation results and alarm information, good primary frequency modulation performance of the generator sets can also be promoted to be maintained, and safe and stable operation of grid frequency is maintained together.