In this chapter, we tackle two important aspects of metaheuristics: their performance and their limitations. Indeed, metaheuristics are stochastic in character, and their rigorous analysis is difficult. We are thus led to computational methods to measure the performance of a metaheuristic and to compare metaheuristics with each other. The empirical analysis makes use of suitably chosen sets of test problems and of well-established metrics and statistics for evaluating such things as the trade-off between the computational effort expended and the solution quality or the success rate of a given metaheuristic on the benchmark problem set. Through the use of examples, we present and discuss tools and results for both discrete and continuous problems. The chapter concludes by describing the fundamental limitations of metaheuristics according to the “no free lunch” theorems.