Cellular Automata are discrete dynamical systems having the ability to generate highly complex behavior starting from a simple initial configuration and set of update rules. However, the discovery of rules exhibiting a high degree of global self-organization for certain tasks is not easily achieved. In this paper, a fast compression based technique is proposed, capable of detecting promising emergent space-time patterns of cellular automata. This information can be used to automatically guide the evolutionary search toward more complex, better performing rules. Results are presented for the most widely studied cellular automata computation problem, the Density Classification Task, where incorporation of the proposed method almost always pushes the search beyond the simple block-expanding rules.