In this paper a metaheuristic approach is proposed for solving the problem of symbolic regression for function approximation. The focus is on developing a method that is easy to implement and can be used to generate initial populations for more advanced metaheuristics. This is achieved by first developing a greedy heuristic which expands (adds terms) generated formulas while increasing the quality of the approximation. This basic algorithm is extended to the Greedy randomized adaptive search procedure (GRASP) by adding randomization and a local search. The local search consists in removing unnecessary terms from the generated formulas. The performed computational experiments show that the GRASP approach, in case of grammars having a limited number of terminal symbols, substantially out performs algorithms based on the artificial bee colony algorithm and ant colony optimization.