This thesis documents the use of optimization on loudspeakers in order to improve its magnitude response and dispersion. The thesis bases its results and methods on the hypothesis that a loudspeaker can be improved by optimizing only the crossover network. A loudspeaker model is created from theories describing a loudspeaker. This model is used as basis for optimization of a loudspeakers magnitude response and dispersion. The optimization method is steepest descent. A reference loudspeaker is constructed from basic methods in order to evaluate the results of optimizing the crossover network. Both the model and optimization algorithm are verified by measurements. The implementations are carried out in Matlab. Simulations and measurements show, that the loudspeaker model and optimization algorithm work satisfying, and the theory verifications show that the theories are valid. When comparing the reference loudspeaker using the standard crossover network and the optimized crossover network it can be concluded, that the magnitude response and the dispersion can be improved by optimizing the crossover network. The optimized loudspeaker has a more flat magnitude response both on-axis and 30 degrees horizontally off-axis. It is concluded that the loudspeaker modelling and optimization are successful.