The invention discloses a power system reactive power optimization method considering discrete variables. The method comprises the following steps: establishing a power system reactive power optimization model which comprises the discrete variables and takes minimum active power loss in the whole network as an objective function; regarding all the undetermined discrete variables in the power system reactive power optimization model as continuous variables, and finding a virtual optimal solution of each discrete variable in the power system reactive power optimization model; calculating sensitivity of the objective function when each discrete variable is at an adjacent virtual gear thereof; calculating the product of gear difference of the discrete variables, of which the absolute value is the maximum, and the corresponding sensitivity, the discrete variables, of which the absolute value is the maximum, being discrete variables to be solved; determining the final result gear of the discrete variables according to the symbol of the sensitivity of the discrete variables to be solved; and after the gear of the discrete variable is determined, regarding the discrete variable as a known variable until all the discrete variables are determined, and obtaining the final optimization result of the discrete variables.