The invention provides a water-turbine generator-set magnetic pull force imbalance fault analysis method based on online data. The method comprises the following steps of according to an unexcited time 1X component vibration vector and an idle condition 1X component vibration vector, calculating vibration and swing variations caused by stator and rotor eccentricity before and after excitation; according to unexcited time 2X-8X component vibration vector and an idle condition 2X-8X component vibration vector, calculating vibration and swing variations generated because a rotor is not round before and after excitation; according to an unexcited time swing gap value and an idle condition swing gap value, calculating a swing gap variation generated because a stator is not round before and after excitation; according to the swing gap variation, calculating an axis offset orientation delta phi ave generated because the stator is not round before and after excitation; if deltaA1x is greater than or equal to deltaA1_1x, determining that a stator and rotor eccentricity fault exists; if deltaA2x_8x is greater than or equal to deltaA1_2x_8x, determining that an out-of-round rotor fault exists; and if delta Save is greater than or equal to deltaS1_ave, determining an out-of-round stator fault exists, wherein an out-of-round most prominent orientation of the stator is located at the delta phi ave. In the invention, according to the online data, a magnetic pull force imbalance fault can be automatically analyzed and determined in real time and an analysis and diagnosis report can be automatically generated.