Changes in major global dynamical phenomena in the Earth's atmosphere are manifested in the time series of atmospheric angular momentum (AAM). as determined directly from meteorological observations and indirectly from geodetic observations of small fluctuations in the rotation of the solid Earth that are proportional to length of day (LOD). AAM fluctuations are intimately linked with energetic processes throughout the whole atmosphere. and also with the stresses at the Earth's surface produced largely by turbulent momentum transport in the oceanic and continental boundary layers and by the action of normal pressure forces on orographic features. A stringent test of any numerical global circulation model (GCM) is therefore provided by a quantitative assessment of its ability to represent AAM fluctuations on all relevant time scales, ranging from months to several years. From monthly data provided by the Atmospheric Model Intercomparison Project (AMIP) of the World Climate Research Programme (WCRP), we have investigated seasonal and interannual fluctuations and the decadal mean in the axial component of AAM in 23 AMIP GCMs over the period 1979-1988. The decadal means are generally well simulated. with the model median value (1.58 x 1026 kg m2 s-1) being only 3.5% larger than the observed mean and with 10 of the models being within 5% of the observed. The seasonal cycle is well reproduced, with the median amplitude of the models' seasonal standard deviations being only 2.4% larger than observed. Half the seasonal amplitudes lie within 15% of the observed and the median correlation found between the observed and model seasonal cycles is 0.95. The dominant seasonal error is an underestimation of AAM during Northern Hemisphere winter associated with errors in the position of subtropical jets. Less robust are the modeled interannual variations, though the median correlation of 0.61 between model simulation and observed AAM is statistically significant. The two El Nino-Southern Oscillation (ENSO) events that occurred during the AMIP decade 1979-1988 have the expected positive AAM anomalies, though the AAM signature of the 1982-1983 event tends to be underestimated, and that of the 1986-1987 event overestimated.