An original theoretical approach for antenna optimization and design is introduced in this paper. This method consists of coupling between the FDTD method applied to the study of microstrip patch antennas and the genetic algorithm (GA) as an optimization tool. A neural network allows the computing time to be deceased by acting as a database, thus avoiding multiple calls to the FDTD code. The principle of this approach is illustrated. (C) 1997 John Wiley & Sons, Inc.