Evolutionary algorithms (EA) are randomized heuristic search methods based on the principles of natural evolution (Banzhaf et al., 1998; Goldberg, 1989; Holland, 1975; Back, 1996; Koza, 1992). If we know how to describe the problem using the terminology of artificial evolution, the EAs are quite easy to apply. Actually, the search for solution(s) is transformed into a search for the best EA setup – a mixture of highly correlated settings and functions (encoding scheme, run-time parameters, fitness (objective) function, selection mechanism. . .). Finding a good EA setup is a problem because EAs are chaotic systems where small variations in initial setup produce large variations in the long-term behavior of the model. A good setup for one problem is mostly unusable for another, although similar problem. Evolutionary algorithm that would be easy to apply in any problem domain would have to be autonomous in a sense that it would regulate its own behavior and would have no need for human intervention (except for the preparation phase, of course). This article discusses the operating principles of such an algorithm and presents its implementation. The Autonomous EA (AEA) is an experiment in the evolution of evolutionary algorithms. It is not much different from existing EAs and the line between the two is sometimes very blurred. Actually, AEA combines known concepts, insights and solutions from EAs, artificial life, chaos theory and complex adaptive systems theory into a new form of evolutionary algorithm. The nomenclature used in different fields is overlapping (for example individual/ solution/object/agent). In AEA the evolving individual represents the solution: a population of individuals (solutions) is evolved in order to find a solution (individual) for the problem at hand. Population is just a limited representation of the vast search space of all possible solutions.