Atmospheric density accounts for the largest source of error for predicting low‐Earth orbiting satellite positions. Relevant applications include catalog maintenance, collision avoidance, and mission design. In this paper we quantify density model errors due to density variability at submodel grid spatial scales. Densities in the 370–450 km height region inferred from accelerometer measurements on CHAMP between 2001 and 2006 are utilized. Results are presented in terms of root‐mean‐square (RMS) amplitudes of the high‐pass‐filtered density residuals with respect to the trend. RMS one‐sigma amplitudes are found to be on the order of 9% (13%) for low (high) geomagnetic activity and to possess distinct seasonal‐latitudinal and local time patterns. The RMS density amplitude patterns suggest a high‐latitude (i.e., auroral) wave source for many of the density perturbations, dissipation for most waves poleward of 40° latitude, and a preference for penetration to lower latitudes during nighttime and during higher levels of geomagnetic activity and lower levels of solar activity. An unknown fraction of the density variability may also originate from gravity wave sources in the lower atmosphere. The RMS amplitude distributions provide statistical constraints on gravity wave contributions from all sources and in a modeling or theoretical context also help to constrain mean heating and acceleration rates due to wave dissipation. In the context of density model errors, our results can be used to calculate an orbit prediction and attitude dynamics error budget. Other engineering applications such as thruster throttling requirements for a drag‐compensated satellite system are also discussed.