This paper presents a comparative study of evolutionary algorithms which are considered to be effective in solving the multilevel lot-sizing problem in material requirement planning (MRP) systems. Three evolutionary algorithms (simulated annealing (SA), particle swarm optimization (PSO) and genetic algorithm (GA)) are provided. For evaluating the performances of algorithms, the distribution of total cost (objective function) and the average computational time are compared. As a result, both GA and PSO have better cost performances with lower average total costs and smaller standard deviations. When the scale of the multilevel lot-sizing problem becomes larger, PSO is of a shorter computational time.In this paper, we compare several evolutionary algorithms for solving the multilevel lot-sizing (MLLS) problem in material requirement planning (MRP) systems. MLLS problem with a time-invariant cost structure is an important decision making process in the manufacturing production systems. For solving such problem there are many researches involving both optimal and heuristic approaches . We provide an overview about it later. The major drawback of the existing approaches is undoubtedly their inability to provide cost-efficient solutions in a reasonable computation time for the realistic size problems. Recently, some researchers [7,27] used evolutionary algorithms to develop a cost efficient solution method that required a low computational effort for solving the MLLS problem. They reported that their algorithms provided high cost effectiveness solutions in a moderate execution time. Then the evolutionary algorithms seem to become useful tools for solving the MLLS problem. On the other hand, because each evolutionary method has its own suitable application scope and constraint condition, it is necessary to evaluate performance for those algorithms in solving the MLLS problem. It is the objective of this paper.The evolutionary algorithms (or memetic algorithms) are a large and diverse class of intelligent algorithms inspired by models of the natural evolution of biological species. The typical examples of the evolutionary algorithms are the simulated annealing (SA) [13] , genetic algorithm (GA) [12] , analog neural networks [23] , ant colony algorithm [6] , particle swarm optimization (PSO) [15] and various hybrid methods [11,18] . In this paper three evolutionary algorithms, SA, PSO and GA, are provided for the performance evaluation. We provide a brief summary of the three evolutionary algorithms in the next section. To assess the output of the algorithms, we compare the algorithms with several lotsizing models, which have different numbers of items