The reciprocally connected GABAergic globus pallidus (GP)-glutamatergic subthalamic nucleus (STN) network is critical for voluntary movement and an important site of dysfunction in movement disorders such as Parkinson's disease. Although the GP is a key determinant of STN activity, correlated GP-STN activity is rare under normal conditions. Here, we define fundamental features of the GP-STN connection that contribute to poorly correlated GP-STN activity. Juxtacellular labeling of single GP neurons in vivo and stereological estimation of the total number of GABAergic GP-STN synapses suggest that the GP-STN connection is surprisingly sparse: single GP neurons maximally contact only 2% of STN neurons and single STN neurons maximally receive input from 2% of GP neurons. However, GP-STN connectivity may be considerably more selective than even these estimates imply. Light and electron microscopic analyses revealed that single GP axons give rise to sparsely distributed terminal clusters, many of which correspond to multiple synapses with individual STN neurons. Application of the minimal stimulation technique in brain slices confirmed that STN neurons receive multi-synaptic unitary inputs and that these inputs largely arise from different sets of GABAergic axons. Finally, the dynamic clamp technique was applied to quantify the impact of GP-STN inputs on STN activity. Small fractions of GP-STN input were sufficiently powerful to inhibit and synchronize the autonomous activity of STN neurons. Together these data are consistent with the conclusion that the rarity of correlated GP-STN activity in vivo is due to the sparsity and selectivity, rather than the potency, of GP-STN synaptic connections.