The sit-to-stand (STS) motion is essential to most activities of daily living, and its difficulty affects the quality of life of many elderly people. Although physical therapy is typically performed to improve mobility, such a service is time-consuming and increasingly limited in number. Supplementing the existing techniques with automated tools should address the growing need of such a service. However, development of effective and reliable tools necessitates the understanding of good STS therapy techniques. We propose to identify quantitative indices that characterize effective STS therapy through the comparison of STS therapy performed by experts and non-experts. First, we measured relevant experimental data from STS motion induced by different therapists on subjects. Second, we used machine learning techniques to extract meaningful features. Finally, we identified the key indices through the analysis of the extracted features. Through such extracted indices, existing techniques can be evaluated and automated rehabilitation tools with measurable performance can be developed.