A method is presented for the real-time estimation of atmospheric density, locally along the trajectory of a low altitude spacecraft. Atmospheric density is estimated simultaneously with six parameters of the spacecraft orbit, and with other parameters that are both observable and unknown. Atmospheric density is derived by estimating local corrections directly to a global atmospheric density model. These corrections are derived, in part, from real-time range and/or Doppler tracking data. They are also derived from F 10.7 and a P measurement values. F 10.7 and a P measurements are used conventionally to drive the global atmospheric density model. But they are also employed by two new stochastic atmospheric density error-models: A baseline error model and a dynamic error model. The baseline error model is derived from the historical record of F 10.7 and a P measurements across multiple eleven year solar cycles. The dynamic error model is an extension to the baseline model, and is derived from current F 10.7 and a P measurements. This provides a new physical connection between the physics of atmospheric density and atmospheric density estimation. Real-time here means that the time-lag for estimation of the local atmospheric density is less than one second following the arrival of new range and/or Doppler tracking data. Atmospheric density estimation is demonstrated with real LEO tracking data acquired during July 2000 - at solar maximum.