In order to develop systems capable of artificial evolution, we need toidentify which systems can produce complex behavior. We present a novelclassification method applicable to any class of deterministic discrete spaceand time dynamical systems. The method is based on classifying the asymptoticbehavior of the average computation time in a given system before entering aloop. We were able to identify a critical region of behavior that correspondsto a phase transition from ordered behavior to chaos across various classes ofdynamical systems. To show that our approach can be applied to many differentcomputational systems, we demonstrate the results of classifying cellularautomata, Turing machines, and random Boolean networks. Further, we use thismethod to classify 2D cellular automata to automatically find those withinteresting, complex dynamics. We believe that our work can be used to design systems in which complexstructures emerge. Also, it can be used to compare various versions of existingattempts to model open-ended evolution (Ray (1991), Ofria et al. (2004),Channon (2006)).