In this work a branch prediction system which utilizes evolutionary techniques is introduced. It allows the predictor to adapt to the executed code and thus to improve its performance on the fly. Experiments with the predictor system were performed and the results display how various parameters can impact its performance on various executed code. It is evident that a one-level predictor can be evolved whose performance is better than comparable predictors of the same class. The dynamic prediction system predicts with a relative high accuracy and outperforms any static predictor of the same class. Categories and Subject Descriptors I.5 [Pattern Recognition]: Miscellaneous General Terms Experimentation Keywords branch prediction, finite automata predictors