The design of multilayer radar absorber (MRA) has become quite popular in the last quarter century, and many studies have been presented to optimize different MRAs. In these studies, the number of layers (NL) has been predetermined, and thicknesses of the layers have been defined in the same parametric sequence. However, according to the type of objective function, the NL and thicknesses may vary, so this initial predetermination may prevent the discovery of the best overall solution. In this study, a double-stage artificial bee colony (DS-ABC) algorithm which evaluates the number and sequence of layers separately from thicknesses is proposed for the optimum design of MRA. Thus, the sequence and thickness of the MRAs are optimized to minimize the reflection coefficient and the total thickness for the desired frequency ranges (2-18 GHz), angles of incidence (0 degrees-40 degrees), and polarizations [transverse electric (TE)-transverse magnetic (TM)]. For the MRA designs, 16 predefined materials and recently published 21 up-to-date materials are used. The designed MRAs are compared with those optimized by other algorithms existing in the literature. The results show that the DS-ABC algorithm is more effective than other conventional metaheuristic search algorithms to find the optimal layer sequence and corresponding thicknesses.