System identification is a method of determining a mathematical relation between variables and terms of a process based on observed input-output data. Model structure selection is one of the important steps in a system identification process. Evolutionary computation (EC) is known to be an effective search and optimization method and in this paper EC is proposed as a model structure selection algorithm. Since EC, like genetic algorithm, relies on randomness and probabilities, it is cumbersome when constraints are present in the search. In this regard, EC requires the incorporation of additional evaluation functions, hence, additional computation time. A deterministic mutation-based algorithm is introduced to overcome this problem. Identification studies using NARX (Nonlinear AutoRegressive with eXogenous input) models employing simulated systems and real plant data are used to demonstrate that the algorithm is able to detect significant variables and terms faster and to select a simpler model structure than other well-known EC methods.