To navigate our complex world, our brains have evolved a sophisticated ability to quickly learn arbitrary rules such as 'stop at red'. Studies in monkeys using a laboratory test of this capacity--conditional association learning--have revealed that frontal lobe structures (including the prefrontal cortex) as well as subcortical nuclei of the basal ganglia are involved in such learning. Neural correlates of associative learning have been observed in both brain regions, but whether or not these regions have unique functions is unclear, as they have typically been studied separately using different tasks. Here we show that during associative learning in monkeys, neural activity in these areas changes at different rates: the striatum (an input structure of the basal ganglia) showed rapid, almost bistable, changes compared with a slower trend in the prefrontal cortex that was more in accordance with slow improvements in behavioural performance. Also, pre-saccadic activity began progressively earlier in the striatum but not in the prefrontal cortex as learning took place. These results support the hypothesis that rewarded associations are first identified by the basal ganglia, the output of which 'trains' slower learning mechanisms in the frontal cortex.