Motor learning was studied in rats (n=8) trained to slow down the forepaw extension while reaching into a narrow tube equipped with an axially moving piston. The photoelectrically monitored velocity of the movement between points P1, P2 and P3 (13.0, 15.1 and 17.3 mm from tube entrance) was evaluated with a laboratory computer which also plotted distributions of the corresponding time intervals and delivered food pellets whenever the time interval between P1 and P2 exceeded a preset threshold value (Group I). The additional condition of attaining P3 was required in Group II. Marked prolongation of the P1-P2 interval required about 20 sessions of 512 reaches in Group I and 27 to 30 sessions in Group II. The rats prolonged the P1-P2 interval in two ways: (a) By generating movements the amplitude of which oscillated around the P2 level and thus exposed the P2 sensor to the slow terminal part of the movement; (b) By true velocity decrease in the middle part (P1-P2) of the movements attaining the level P3. It is concluded that reaching is a highly stereotyped instrumental reaction, the pattern of which is not easily modified by operant conditioning. Substantial slowing (2 times) of the ballistic extension phase of long reaches probably reflects programmed coactivation of the antagonist.