Many optimization problems (from academia or industry) require the use of a metaheuristic to find a satisfying solution in a reasonable amount of time, even if optimality is not guaranteed. Metaheuristics can be roughly partitioned in two groups: local search methods (e.g., simulated annealing, tabu search, and variable neighborhood search) and population based algorithms (e.g., genetic algorithms, ant colonies, and scatter search). The reader is assumed to be familiar with the most popular metaheuristics. Even if there exist convergence theorems for some metaheuristics, they usually do not help to develop an efficient metaheuristic. The goal of this paper is to propose general rules which are useful when designing metaheuristics in order to produce good performance according to several criteria, independently of the class of metaheuristics employed. The discussion is illustrated for three well-known optimization problems: graph coloring, vehicle routing and job-shop scheduling.