A previous study (Pardini C., Anselmo L, Moe K., Moe M.M., Drag and energy accommodation coefficients during sunspot maximum, Adv. Space Res., 2009, doi:10.1016/j.asr.2009.08.034), including ten satellites with altitudes between 200 and 630 km, has yielded values for the energy accommodation coefficient as well as for the physical drag coefficient as a function of height during solar maximum conditions. The results are consistent with the altitude and solar cycle variation of atomic oxygen, which is known to be adsorbed on satellite surfaces, affecting both the energy accommodation and angular distribution of the reemitted molecules. Taking advantage of these results, an investigation of thermospheric density model biases at sunspot maximum became possible using the recently upgraded CDFIT software code. Specif-ically developed at ISTI/CNR, CDFIT is used to fit the observed satellite semi-major axis decay. All the relevant orbital perturbations are considered and several atmospheric density models have been implemented over the years, including JR-71, MSISE-90, NRLMSISE-00, GOST2004 and JB2006. For this analysis we reused the satellites Cosmos 2265 and Cosmos 2332 (altitude: 275 km), SNOE (altitude: 480 km), and Clementine (altitude: 630 km), spanning the last solar cycle maximum (October 1999 -January 2003). For each satellite, and for each of the above men-tioned atmospheric density models, the fitted drag coefficient was obtained with CDFIT, using the observed orbital decay, and then compared with the corresponding physical drag coefficient estimated in the previous study (Pardini et al., 2009). It was consequently possible to derive the average density biases of the thermospheric models during the considered time span. The average results obtained for the last sunspot maximum can be summarized as follows (the sign "+" means that the atmospheric density is overestimated by the model, while the sign "-" means that the atmospheric density is underestimated): Cosmos 2265 (275 km) density biases: JR-71: +10.6%; MSISE-90: +10.1%; NRLMSISE-00: +8.8%; GOST2004: +10.6%; JB2006: +11.8%; Cosmos 2332 (275 km) density biases: JR-71: +11.0%; MSISE-90: +10.4%; NRLMSISE-00: +8.9%; GOST2004: +9.7%; JB2006: +11.8%; SNOE (480 km) density biases: JR-71: +17.5%; MSISE-90: +10.0%; NRLMSISE-00: +6.4%; GOST2004: +8.2%; JB2006: +14.4%; Clementine (630 km) density biases: JR-71: +14.2%; MSISE-90: +5.7%; NRLMSISE-00: -0.1%; GOST2004: -10.9%; JB2006: +14.6%. Below 500 km all the models overestimated the atmospheric density: the biases were comparable at the lowest altitude considered (275 km), while a wider spread of the values was obtained at 480 km. A different behavior was instead observed at 630 km, where MSISE-90 and NRLMSISE-00 were found to be affected by significantly smaller biases and GOST2004 exhibited a negative bias.