Assembly lines are the most widely used systems for industrial mass production. A main objective in such a system is to ensure a workload balancing among its workstations and optimize it at the operational level. However, this balancing is affected by various disturbances which induce delays and then generate additional costs and deteriorate the performance of the assembly line. To remedy the negative effects of such disturbances, methods allowing real-time rebalancing are needed. The problem is known as the dynamic rebalancing of assembly lines. This work proposes a comparative study of three metaheuristics performances in solving this problem, namely: Iterated Local Search, Genetic Algorithm and Filters Beam Search-Ant Colony Optimization. The choice of these metaheuristics is motivated by their reputation for quickly and efficiently solving assembly line balancing problems. An exact method, whose performance is compared to the three selected metaheuristics, is also considered. The four approaches are applied to instances of industrial size and complexity known in the assembly line balancing literature. This benchmark data set guarantees coverage of almost all cases that an industrial could encounter. Obtained results showed the metaheuristics efficiency in solving large instances and that the exact method is recommended for small ones. Efficiency is measured here in terms of resolution speed (few seconds are required) and the quality of returned rebalancing solution. A rebalancing solution is of good quality if its cycle time is less than or equal to the initial line takt time.