In this work, a novel computer‐aided design methodology for probe‐fed, cavity‐backed, stacked microstrip patch antennas is proposed. The methodology incorporates the rigor of a numerical technique, such as finite element methods, which, in turn, makes use of a newly developed procedure (multipurpose admittance matrices) to carry out a full‐wave analysis in a given structure in spite of certain physical shapes and dimensions not yet being established. With the aid of this technique, we form a training set for a neural network, whose output is the desired response of the antenna according to the value of design parameters. Last, taking advantage of this neural network, we perform a global optimization through a genetic algorithm or simulated annealing to obtain a final design. The proposed methodology is validated through a real design whose numerical results are compared with measurements with good agreement. © 2008 Wiley Periodicals, Inc. Microwave Opt Technol Lett 50: 2411–2416, 2008; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/mop.23670