GPS precise positioning depends on the right determination of phase ambiguity vector. Generally there are three methods for integer ambiguity resolution, including direct rounding to integer method, searching method and ambiguity function method. Direct rounding to integer method is the fastest algorithm among the three methods, but it requires highly precise approximation of the ambiguities, which usually can be only obtained through long term of measurement. The efficiency (speed) of searching method has a very close relationship to the precision of real-valued ambiguity solution, and usually this method is very slow, especially for the data which comes from short time measuring. The efficiency of ambiguity function method is determined by the accuracy of approximation position and the volume of the cube centered at the initial position. The original ambiguities are transformed into another space by a modified ambiguity transformation algorithm purposed by this paper, which is the most effective one compared to the others. In the new ambiguity space, new ambiguities are far more precise than their original partners. Then we compute the successful probability of direct rounding to integer method both in original space and new space. If the success probability is larger than a given value, then direct rounding to integer method is used to obtain the true ambiguities, otherwise go to a subroutine for the searching method.