The mechanisms of decision making are generally thought to be under the control of a set of cortico-subcortical loops. There are several parallel functional loops through the basal ganglia connecting back to distinct areas of cortex, process- ing different modalities of decision making, including motor, cognitive and limbic. Due to convergence and divergence within the network, these loops cannot be completely segregated. We use these circuit properties to develop a connec- tionist model at a spiking neuron level of description, relying on the bases of the recently published Guthrie’s model1. This model is applied to a decision making task that has been studied in primates. The electrophysiological results of this work showed that the decision process is done in two successive steps. In this task, the animals are trained to associate reward values to target shapes in order to maximize their gain2. To develop this model, we use two parallel loops, each of which performs decision making based on interactions between positive and negative feedback pathways within the loop. The loops communicate via partially convergent and divergent connections in one specific area. This model is tested to perform a two level decision making as in primates. The whole system is instantiated using leaky integrate- and-fire neurons and its architecture relies on commonly accepted data regarding the complex functional connectivity description between basal ganglia, cortex and thalamus. A bottom-up approach of the basal ganglia model in which the learning of optimum decision making is thus developed. This capability will emerge from the closed-loop interaction between the neural circuitry and its sensory-motor interface. This model will allow us (i) to avoid the arbitrary choice of a pre-existing machine-learning derivative model and also provides (ii) to have the possibility to investigate the cell-scale mechanisms impact on the whole model capacities.