This study investigated the process that contributes to the decay of short-term motor memory regarding force reproduction. Participants performed tonic flexion of the right index finger with the target force feedback (criterion phase) and reproduced this force level without feedback 3, 10, 30, or 60 s after the end of the criterion phase (recall phase). The constant error for force reproduction was significantly greater than zero, indicating that information about the somatosensation and/or motor command in the criterion phase is positively biased. Constant and absolute errors were not influenced by the retention interval, indicating that neither bias nor error represents the decay of short-term motor memory over time. Variable error, defined as SD of bias (force in the recall phase minus that in the criterion phase), increased as the retention interval increased. This indicates that the decay of short-term motor memory is represented by the increase in inconsistency of memory bias among the trials. The correlation coefficient of the force between the criterion and recall phases with 3-s retention interval was greater than that with longer intervals. This is explained by the view that the contribution of the information of the practiced force to the force reproduction process is great within 3 s after the end of the practice, but the additional contribution of the noise information becomes greater after this time, causing lesser relative contribution of the information of the practiced force to the force reproduction process.