Gene Expression Programming (GEP) is a new technique of Genetic Programming (GP) that implements a linear genotype representation. It uses fixed-length chromosomes to represent expression trees of different shapes and sizes, which results in unconstrained search of the genome space while still ensuring validity of the program’s output. However, GEP has some difficulty in discovering suitable function structures because the genetic operators are more disruptive than traditional tree-based GP. One possible remedy is to specifically assist the algorithm in discovering useful numeric constants. In this paper, the effectiveness of several constant creation techniques for GEP has been investigated through two symbolic regression benchmark problems. Our experimental results show that constant creation methods applied to the whole population for selected generations perform better than methods that are applied only to the best individuals. The proposed tune-up process for the entire population can significantly improve the average fitness of the best solutions.