A cost function was used to compare the calculation pattern (associated to the chromosome encoded by GA) with the degraded pattern at the samples measured from the far field radiation pattern. Genetic algorithm was used to minimize the cost function to locate the failed elements. However, it was found that when using GA to diagnose the failed elements in an array, this method could not accurately distinguish symmetrical positions if the excitation of the arrays were symmetrical. In order to solve this ambiguity problem, an asymmetry factor was introduced in the mentioned diagnosis method to the excitation distribution (amplitude or phase) of the arrays artificially. This improved method was exploited to a linear array to validate the method. From the comparison between the symmetrical and asymmetrical excitation distributions of the arrays, it was showed that the improved method was applicable. And weight coefficients were introduced in the cost function in order to get better results.