In Parkinsonism, subthalamic nucleus (STN) neurons and two types of external globus pallidus (GP) neuron inappropriately synchronise their firing in time with slow (∼1 Hz) or beta (13-30 Hz) oscillations in cortex. We recorded the activities of STN, Type-I GP (GP-TI) and Type-A GP (GP-TA) neurons in anaesthetised Parkinsonian rats during such oscillations to constrain a series of computational models that systematically explored the effective connections and physiological parameters underlying neuronal rhythmic firing and phase preferences in vivo. The best candidate model, identified with a genetic algorithm optimising accuracy/complexity measures, faithfully reproduced experimental data and predicted that the effective connections of GP-TI and GP-TA neurons are quantitatively different. Estimated inhibitory connections from striatum were much stronger to GP-TI neurons than to GP-TA neurons, whereas excitatory connections from thalamus were much stronger to GP-TA and STN neurons than to GP-TI neurons. Reciprocal connections between GP-TI and STN neurons were matched in weight, but those between GP-TA and STN neurons were not; only GP-TI neurons sent substantial connections back to STN. Different connection weights between and within the two types of GP neuron were also evident. Adding to connection differences, GP-TA and GP-TI neurons were predicted to have disparate intrinsic physiological properties, reflected in distinct autonomous firing rates. Our results elucidate potential substrates of GP functional dichotomy, and emphasise that rhythmic inputs from striatum, thalamus and cortex are important for setting activity in the STN-GP network during Parkinsonian beta oscillations, suggesting they arise from interactions between most nodes of basal ganglia-thalamocortical circuits.