First-order theorem provers have to search for proofs in an infinitespace of possible derivations. Proof search heuristics play a vitalrole for the practical performance of these systems. In the currentgeneration of saturation-based theorem provers like SPASS, E,Vampire or Prover~9, one of the most important decisions is theselection of the next clause to process with the given clausealgorithms. Provers offer a wide variety of basic clause evaluationfunctions, which can often be parameterized and combined in manydifferent ways. Finding good strategies is usually left to the usersor developers, often backed by large-scale experimentalevaluations. We describe a way to automatize this process usinggenetic algorithms, evaluating a population of different strategieson a test set, and applying mutation and crossover operators to goodstrategies to create the next generation. We describe the design andexperimental set-up, and report on first promising results.