Abstract. With the help of four years (2002–2005) of CHAMP accelerometer data we have investigated the dependence of low and mid latitude thermospheric density on the merging electric field, E_m, during major magnetic storms. Altogether 30 intensive storm events (Dst_min<−100 nT) are chosen for a statistical study. In order to achieve a good correlation E_m is preconditioned. Contrary to general opinion, E_m has to be applied without saturation effect in order to obtain good results for magnetic storms of all activity levels. The memory effect of the thermosphere is accounted for by a weighted integration of E_m over the past 3 h. In addition, a lag time of the mass density response to solar wind input of 0 to 4.5 h depending on latitude and local time is considered. A linear model using the preconditioned E_m as main controlling parameter for predicting mass density changes during magnetic storms is developed: ρ=0.5 E_m + ρ_amb, where ρ_amb is based on the mean density during the quiet day before the storm. We show that this simple relation predicts all storm-induced mass density variations at CHAMP altitude fairly well especially if orbital averages are considered.