The temporal variability and forcing of global atmospheric angular momentum (AAM) is studied using a three-component Markov model derived from observed statistics of global AAM and the global torques. The model consists of stochastic forcing by the mountain (&tgr;M) and friction (&tgr;F*) torque plus a pervasive negative feedback on AAM by the friction torque. AAM anomalies are damped at a 30-day timescale and forced by torques having 1.5-day (&tgr;M) and 6-day (&tgr;F*) decorrelation timescales. A large portion of the intraseasonal variance and covariance of AAM, &tgr;M, and &tgr;F is accounted for by the Markov model. Differences between the modeled and the observed covariances are maximized in the 10- to 90-day band and account for 10--30% of the variance when using data not stratified by season. An especially prominent deviation from the Markov model is the oscillatory forcing of AAM by the frictional torque at 30- to 60-day periods. Additionally, there is greater coherent variance between &tgr;F and &tgr;M across the entire 10- to 90-day band, with the frictional torque leading the mountain torque. This feedback between the global torques results from physical processes not represented in the Markov model. The synoptic characteristics of the stochastic mountain and frictional torques and of the oscillatory Madden-Julian Oscillation are described. ¿ 2000 American Geophysical Union