Activity of striatal projection neurons represents reward expectation, and that dopamine neurons code reward expectation error and an incentive to work for reward, the basal ganglia have been supposed to be involved in adaptive reward-based action mechanisms. However, it remains to be revealed that how the tonically active neurons (TANs), participate in the adaptive reward-based action mechanisms. Thus, we studied the activity of TANs during learning of reward-based decision-making task. Monkey learned to depress a hold button after illumination of the button (start cue), and choose one of 3 target buttons by trial-and-error. If monkey chose a correct button, high-beep occurred and water was given. But low-beep indicate no reward (incorrect choices). The correct choice probabilities were 30% in first choice (N1), 50% at second choice (N2), around 90% at third choice (N3) and at repetition trials (R1). The monkey, first, learned 2 target task, then learned 3 target task. TANs responded to start cue and beep by suppression and/or facilitation of tonic discharges. Beep responses were similar in all trial types during the initial stage. But, through behavioral learning, responses became strong in N1 and N2 trials but very small in R1 trials. In contrast, start cue responses were stronger in N1 and R1 trials after learning. Thus, reward probability of decision strongly influenced activity of TANs, different from dopamine neurons. These results suggested that TANs modify their activity through behavioral learning of reward-based decision-making strategy. <b>[Jpn J Physiol 55 Suppl:S185 (2005)]</b>