In order to improve the low optimization efficiency and the premature convergence of genetic algorithms (GA), a multipopulation agent cogenetic algorithm with a chainlike agent structure (MPAGA) was developed. This algorithm adopted a multipopulation parallel searching mode, a chainlike agent structure, dynamic neighborhood competition, and an orthogonal crossover strategy. Three functions were used to test this algorithm. The experimental results show that MPAGA obtains higher optimization precision and converges to the domain close to global optima with higher speeds than other improved GAs.