In this paper we present a framework for modelling Simple and Parallel Evolutionary Algorithm implementations as Co-operating Populations. Using this framework, a method called Co-operating Populations with Different Evolution Behaviours (CoPDEB), for generalizing and improving the performance of Parallel Evolutionary Algorithms (PEAs) is also presented. The main idea of CoPDEB is to maintain a number of populations exhibiting different evolution behaviours. CoPDEB was tested on three problems (the optimization of a real function, the TSP problem and the problem of training a Recurrent Artificial Neural Network), and appears to significantly increase the problemsolving capabilities over PEAs with the same evolution behaviour on each population. This paper also studies the effect of the migration rate (Epoch) and the population size on the performance of both PEAs and CoPDEB.