The use of active compliance to realize an error-corrective assembly is addressed. A robot operating under active compliance-control may be viewed as continuously computing motions from forces. A compliance matrix specific to the shapes of the parts to be mated is constructed. During execution of the assembly operation, this compliance matrix efficiently maps measured forces into appropriate corrective motions. Essentially, the plan is to program, in advance, the 9 (in 2-space) or 36 (in 3-space) elements of a compliance matrix so that the robot will respond correctly to any error which may occur during the assembly operation. Three desirable criteria are proposed for the motion of an error-corrective manipulator. The three criteria become mathematical requirements on the compliance matrix to be designed. It is shown how the compliance matrix which best fulfills these criteria can be constructed.