We statistically controlled fitness (1.5-mile run), outcome-expectancy values, and perceived barriers for physical activity and then compared internal health locus of control (IHLOC) with internal exercise locus of control (IEXLOC) for predicting the physical activity of college students (N = 84). Prospective observations of self-reported free-living physical activity (seven-day recall) and supervised running (time x distance) were made at 2-, 5-, and 9-week intervals. IHLOC predicted seven-day recall at Week 2 (beta = .19) and Week 5 (beta = .36) (increase in adjusted R2 ranged from .05 to .12, p less than .05). Consistent with theory, the prediction was not reproducible when generalized outcome-expectancy value rather than outcome-expectancy value for health was assessed. IHLOC was unrelated to supervised running. IEXLOC was unrelated to activity in all analyses, even though exercise-specific measures of outcome-expectancy values (beta s = .20) and perceived barriers (beta s = .27-.32) predicted both seven-day recall and supervised running (p less than .05). IHLOC also discriminated (p less than .05) high active and low active subjects when criterion groups were formed from population norms on seven-day recall (greater than or less than 280 kcal.kg-1.week-1). Results indicate that tests of decision theories that include locus of control measures specific to health or exercise must adjust for fitness, perceived barriers to physical activity, and behaviorally relevant outcome-expectancy values of physical activity when studying college students. The construct validity of exercise locus of control remains uncertain for college students.