Simulations of natural variability by two GCMs are examined. One GCM i
s a sector model, allowing relatively rapid integration without simpli
fication of the model physics. which would potentially exclude mechani
sms of variability. Two mechanisms are found in which tropical surface
temperature and SST vary on interannual and longer timescales. Both a
re related to changes in cloud cover that modulate SST through the sur
face radiative flux. Over the equatorial ocean, SST and surface temper
ature vary on an interannual timescale, which is determined by the mag
nitude of the associated cloud cover anomalies. Over the subtropical o
cean, variations in low cloud cover drive SST variations. In the secto
r model, the variability has no preferred timescale, but instead is ch
aracterized by a ''red'' spectrum with increasing power at longer peri
ods. In the terrestrial GCM, SST variability associated with low cloud
anomalies has a decadal timescale and is the dominant form of global
temperature variability. Both GCMs are coupled to a mixed layer ocean
model, where dynamical heat transports are prescribed, thus filtering
out ENSO and thermohaline circulation variability. The occurrence of v
ariability in the absence of dynamical ocean feedbacks suggests that c
limatic variability on long timescales can arise from atmospheric proc
esses alone.