The transducer under consideration is a planar 2D array transducer working at 3.5 MHz. The transducer is composed of 17 by 17 piezoelectric elements separated by major and minor kerfs, a thick backing layer, two front impedance matching layers, and an acoustic lens. Three-dimensional finite element models of the transducer were constructed using the commercial finite element analysis (FEA) code PZFlex. The performance of an acoustic transducer is determined by the effects of many structural variables, and in most cases the influences of these variables are not linearly independent of each other. In this paper, through the FEA, the performance variation of the 2D array transducer was investigated by considering all the cross-coupled effects of its structural variables such as the thicknesses of the matching layers, and the width and depth of the major and minor kerfs. Based on the analysis, new structures of the 2D array were proposed to reduce the cross talk. Through statistical multiple regression analysis of the FEA results, the functional forms of the cross talk level, pulse-echo sensitivity, and acceptance angle of the 2D array were derived in terms of the structural variables of the new structures. Then, by applying the constrained optimization technique, genetic algorithm, to the derived functions, the optimal combination of the structural variables, i.e. optimal structure of the transducer, was determined to provide the lowest cross talk level and highest pulse-echo sensitivity while preserving a desired acceptance angle at the center frequency of 3.5 MHz.