Accelerometer-derived neutral mass density (NMD) is an important quantity describing the variability of the upper atmosphere. NMD is widely used to calibrate and validate some models used for satellite orbit determination and prediction. Quantifying the true NMD is nearly impossible due to, among others, the lack of simultaneous in-situ measurements for cross-validation and the incomplete characterization of the uncertainties of these NMD products. This study investigates the error distribution of three different accelerometer-derived NMD products from the CHAMP satellite mission during time periods of both high and low solar activity. Using a multimodel ensemble comprised of both physical and empirical models, the study characterizes the error variance of the NMD. The strategies employed here may be useful and applicable to other space missions spanning over longer time periods. The results show considerable differences among the three CHAMP data sets and also reveal a pronounced latitude dependence in their error distributions. The median error standard deviation of CHAMP NMD is smaller during time periods of high solar activity (11.0%) than during periods of low solar activity (13.1%). The results indicate that the method of processing the accelerometer data has a significant impact on the uncertainty estimates of the different CHAMP NMD products.