A robot can only reach discrete locations within its workspace due to its finite encoder counts. Actual position and orientation reached by end-effectors are therefore the result of roundoff in the encoder counts. Traditionally the roundoff is carried out at the end of inverse kinematics, which produces the propagation and accumulation of the roundoff errors (except for Cartesian type of robot). We proposed a sequential roundoff procedure during the inverse kinematics process to compensate the roundoff errors. Numerical analysis shows that this method can lead to significant improvement in the positioning accuracy. Our study also shows that the sequence with which the inverse is carried out affects the improvement. Selection of sequence for inverse kinematics is discussed.