In order to cope with the frequent unpredictable changes that may occur in manufacturing systems, and to optimize given performance criteria, the assignment of workers can be decided online in a dynamic manner, whenever the worker is released. Several studies, in the ergonomics literature, have shown that individuals' performances decrease because of their fatigue in work. In manufacturing context, the workers’ fatigue impacts the task durations. Therefore, we propose to solve the online workers assignment problem through a heuristic, which takes this workers' fatigue into consideration, so as to minimize the mean flowtime of jobs. This approach suggests computing more realistic task duration in accordance with the worker's fatigue and it uses multi-criteria analysis in order to find a compromise to favor short durations and to avoid congestions. The Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) method is used to select where to assign the worker. A learning process through simulation optimization is used to adapt the weights, in TOPSIS, to better fit with the system characteristics. The approach is illustrated with a simulated Job-Shop system. Experimental results comparing our approach with the rule Shortest Processing Time (SPT), which is known as efficient on the mean flowtime, show the effectiveness of the heuristic.