Humans and animals learn the value of stimuli and actions through experience to make advantageous decisions. The cortico-basal ganglia circuits are thought to be centrally involved in such processes, with the key finding that the activity of midbrain dopamine (DA) neurons appears to represent the temporal-difference (TD) reward prediction error (RPE), the difference between reward obtained or expected to be obtained and reward that had been expected to be obtained. How DA neurons compute TD-RPE remains elusive, and we have proposed a possible mechanism for it (Morita et al., 2012, Trends Neurosci; 2013, J Neurosci) named the cortico-striatal TD hypothesis: (1) two subtypes of corticostriatal neurons represent 'current state/action' and 'previous state/action' by virtue of intracortical circuit architecture; (2) 'values' of those states/actions are computed by two subtypes of striatal projection neurons due to anatomically suggested preferential activation of them by the two corticostriatal neuron subtypes; and (3) the difference of those values, which is the core of TD-RPE, is computed in DA neurons through presumable net positive and negative impacts from the two striatal neuron subtypes. To our pleasure, this hypothesis has been recognized in the community (e.g., Shepherd, 2013, Nat Rev Neurosci). However, a recent optogenetic study (Kress et al., 2013, Nat Neurosci) has reported results that challenge the validity of one of its major assumptions. Specifically, the optogenetic study did not find evidence for the biased activation of the different striatal cell types by the different cortical cell types that we assumed based on the anatomical suggestions; even partially opposite bias was instead observed. Nonetheless, subsequent computational work (Morita, in press, J Neurophysiol) suggests, through fitting of past physiological results by a model constrained by the optogenetic results, that our assumption could still be valid for repetitive inputs, rather than for brief inputs examined in the optogenetic study, by virtue of anatomical connection biases and synapse type-dependent short-term plasticity. In this talk, I will introduce the works mentioned above, and discuss future directions.