To examine the feasibility of the long-term production of the human granulocyte colony stimulating factor (hG-CSF) using the L-arabinose promoter system of Escherichia coli, flask relay culture and cyclic fed-batch culture were performed. In the flask relay culture, it was found that the plasmid was maintained stably up to about 170 generations in arm uninduced condition, whereby the cells could also maintain the capability of expressing hG-CSF upon induction. However, in an induced condition, the structural damage of the plasmid occurred after about 100 generations, and thereafter the hG-CSF expression decreased gradually. In both cases, it was observed that the plasmid and the hG-CSF expression were maintained stably up to at least 100 generations. In contrast, in the cyclic fed-batch culture, segregational plasmid instability was observed within about 4 generations after induction, even though the cell growth and hG-CSF production reached their maximum values, 78.0 g/l of dry cell weight and 7.0 g/l of hG-CSF, respectively. It would appear that, when compared to the flask relay culture, the high-cell density and high-level expression of hG-CSF in the cyclic fed-batch culture led to the segregational plasmid instability; in other words, a severe metabolic burden existed on the cells due to the high-level expression of hG-CSF, Accordingly, based on these long-term cultures, the segregational and structural plasmid instability was observed and a strategy to overcome such problems could be designed.