AGAPE, an environment for the implementation of parallel evolutionary algorithms on the APE100/Quadrics architecture, is presented. AGAPE is a flexible tool for users to solve numerical optimization problems; in addition researchers can use AGAPE for the study of new genetic operators, selection functions and migration strategies. A version of AGAPE dedicated to the design of neural networks of arbitrary kind and activation function is described and an original comparison among different evolutionary algorithms to learn neural networks is reported. The simulation results show that the coding scheme adopted by AGAPE leads to the best learning rate on Rumelhart's problems. The paper also includes two reviews on parallel genetic algorithms and on evolution programs for the optimization of neural networks.