A method of building optical proximity correction model comprises the following steps of collecting optical proximity correction data to a database from processes, determining an initial value X0, an end value Xn and a fixed step size h, calculating initial value slope coefficient according to the initial value X0, obtaining an initial optimal step size Y0 according to the initial value slope coefficient, obtaining a first additive value X1 according to the initial value X0 and the initial optimal step size Y0, repeating the steps, working out a second additive valueX2, a third additive value X3,..., till the end value Xn, fitting the initial value X0, the first additive value X1, the second additive value X2, the third additive value X3,..., till an n-1 additive value and the end value Xn. Through the above steps, as step size values can be regulated according to the slope coefficient worked out, regulation step number can be reduced and simulation time is shortened.