We used the uncontrolled manifold (UCM) approach to study the synergy formation during learning an unusual multi-finger task. The subjects produced accurate force ramps with challenging sets of four fingers (two per hand). We tested hypotheses on stabilization of the contributions of subsets of effectors to the task force ( F(TASK)) and to the moment in the frontal plane (force-stabilization and moment-stabilization, respectively). Force signals were used to compute magnitudes of hypothetical independent signals, modes. The variance of the mode magnitudes across repetitions of the task was partitioned into two components, within the UCM ( V(UCM)), which did not affect the average value of a selected performance variable (force or moment), and orthogonal to the UCM ( V(ORT)), which affected the variable. Prior to practice, subjects showed high error indices and failed to show stabilization of each hand's contribution to F(TASK) ( V(ORT)> or = V(UCM)), while the pronation-supination moment was stabilized by the fingers of each hand ( V(ORT)< V(UCM)). The total forces produced by each of the two hands showed negative covariation across trials, which supported the force-stabilization hypothesis but not moment-stabilization hypothesis. Both force-stabilization and moment-stabilization hypotheses were supported by analysis of mode magnitudes to all eight fingers. Over 2 days of practice, the performance of the subjects improved considerably. This was accompanied by the emergence of within-a-hand force-stabilization for each of the two hands without deterioration of moment-stabilization. Quantitatively better within-a-hand force-stabilization was seen in male subjects as compared to females throughout the course of the experiment. Force-stabilization by all eight fingers improved quantitatively with practice. Practice also resulted in higher finger forces in maximal force production (MVC) trials and higher forces produced by unintended fingers in single-finger MVC trials (higher enslaving). We conclude that the UCM approach allows quantifying changes in the coordination of effectors during practice, and offers insights into the microstructure of this coordination with respect to different performance variables and different subsets of effectors. The approach can be used to test whether new synergies emerge in the process of practice.