To assess a method of predicting the final postoperative flexion in individual cases after total knee arthroplasty, 364 primary posterior cruciate-retaining total knee arthroplasties were reviewed retrospectively. The knees were subdivided into three preoperative flexion groups--I: poor motion (0 degrees to 85 degrees), II: intermediate motion (90 degrees to 110 degrees), and III: good motion (115 degrees to 140 degrees). There were 302 cases of osteoarthritis and 62 rheumatoid knees (12 juvenile rheumatoid). Correlation was made between preoperative; intraoperative, and postoperative (minimum 2-year follow-up) passive knee flexion for individuals. Intraoperative flexion against gravity was measured after capsular closure by passively flexing the patient's hip 90 degrees and allowing the weight of the lower leg to flex the knee joint. The overall mean value of postoperative flexion for all three groups was similar to preoperative and intraoperative flexion in both osteoarthritis and rheumatoid arthritis. In the poor motion group (I), postoperative flexion (103 degrees) was increased over preoperative flexion (84 degrees) but similar to intraoperative flexion (104 degrees). In the intermediate group (II), postoperative flexion (110 degrees) was similar to both the preoperative flexion (108 degrees) and intraoperative flexion (110 degrees). In the good group (III), postoperative flexion (119 degrees) tended to be less than preoperative flexion (123 degrees) and more than intraoperative flexion (116 degrees), but the differences were not statistically significant. When comparing preoperative and intraoperative flexion to postoperative flexion for individual cases, 55% of knees had postoperative flexion +/-10 degrees of their preoperative value, while 97% of knees had postoperative flexion +/-10 degrees of their intraoperative value. This study indicates that the final postoperative mean flexion for a group of patients with poor preoperative flexion (<85 degrees) and for individual cases (regardless of their preoperative mobility) can best be predicted by intraoperative flexion against gravity rather than by a preoperative value.