Four kinds of solar indices F 10.7, E 10.7, S 10, Mg 10 and four thermospheric neutral density models, i.e., CIRA72, DTM94, NRLMSISE00 and JB2006, are discussed. The CHAMP accelerometer data are used to calculate thermospheric total mass density. Based on the comparison of the model densities with CHAMP observations, the effects of various indices on the model accuracy are detected. It is found that under quiet and moderate solar conditions (F 10.7<160), all of the models’ errors are reduced about 15% by using E 10.7 instead of F 10.7, while under active solar conditions (F 10.7>200) the error’s standard deviation using E 10.7 increases quickly and causes the models’ accuracy to fall down. With regard to S 10, Mg 10, their effects under quiet solar conditions are inconspicuous. If under active solar conditions, they can reduce the model error’s standard deviation by 5%–10%, implying that S 10, Mg 10 make the model error more stable. The JB2006 model, which was constructed by multi-solar-index (F 10.7, S 10, Mg 10), is compared with DTM94 and NRLMSISE00 based on single-solar-index. It is found that JB2006’s accuracy is better than DTM94’s, and is close to NRLMSISE00 under the quiet solar condition. During the solar burst occurring on October 26, 2003, JB2006 has been in best agreement with CHAMP observations. All in all, the new indices may improve thermospheric density models’ accuracy under some special conditions. Concretely, E 10.7 may reduce the average error of models and S 10, Mg 10 may prevent the error’s divergence.