Signal detection methodology was used to identify the best combination of predictors of long-term exercise adherence in 269 healthy, initially sedentary adults ages 50-65 years. Less educated individuals who were assigned to supervised home-based exercise of either higher or lower intensity and who were less stressed and less fit at baseline than other individuals had the greatest probability of successful adherence by the 2nd year. Overweight individuals assigned to a group-based exercise program were the least likely to be successful 2 years later. Predictors of short-term (1-year) adherence were generally similar to predictors of 2-year adherence. Signal detection analysis may be useful for identifying subgroups of people at risk for underadherence who subsequently might be targeted for intervention.