This investigation examined predictors of compliance with exercise therapy in a clinical trial involving older adults with knee osteoarthritis (OA). The study sample was partitioned into tertiles by level of compliance to determine its effect on several clinical outcome measures in the trial (i.e., knee pain, difficulty with activities of daily living, and performance-related disability). The participants (N = 439) first completed all baseline assessments and were then randomly assigned to one of three treatment conditions: health education control, aerobic exercise, or resistance exercise. The two exercise treatments involved a 3-month center-based phase and a 15-month home-based phase. Variables in five categories (i.e., demographic, fitness, health-related quality of life, performance-related disability, and prior exercise behavior) were entered as predictors of attendance and time spent exercising during each session for three different periods of time across the course of the study. Results of these analyses revealed that it was possible to explain more variance for time spent exercising (approximately 40%) during the first 3 months than for attendance (approximately 10%). Furthermore, once participants completed the first 3 months of their training, prior behavior was the strongest predictor of exercise compliance. In most cases, the regression models accounted anywhere from 26 to 46% of the variance in attendance or time spent exercising (7 of the 8 P values < 0.01). In general, demographic, fitness, psychosocial, and disability-related measures did not predict compliance with any consistency across the various phases of the trial. Analysis of the dose-response data suggest that, in the use of aerobic exercise to deter disability in older people with knee OA, consideration should be given to prescribing frequent bouts of activity (at least 3 times each week) of moderate duration (approximately 35 min).