The aims of this study were to explore the relationships between goal specificity, goal difficulty and performance, and to determine if setting unrealistic goals would produce decreases in performance. The subjects were high school students from 15 schools and an attempt was made to control for the effects of social comparison. The schools were randomly assigned to one of 15 conditions representing five levels of goal conditions-namely, 'do' (no goals), 'do your best', 'improve by 10%' (easy), 'improve by 20%' (difficult/ realistic) and 'improve by 40%' (improbable/unattainable)-and three levels of practice duration (4, 6 and 8 weeks). This design consisted of nesting goal difficulty within practice duration, which enabled an examination of the goal specificity and goal attainability/difficulty hypotheses proposed by Locke and Latham (1985). A 5 x 3 factorial ANCOVA was applied to the post-baseline sit-up gain scores. The results indicated that all specific groups performed better than all non-specific groups. In addition, across practice durations the difficult/realistic group exhibited the greatest increase in performance, followed by the easy group. The performance gains of the improbable/unattainable group were substantially less compared with the difficult/ realistic group after 4 and 6 weeks, but not after 8 weeks of practice. These results are in line with both the goal specificity and goal difficulty hypotheses derived from the application of Locke's goal-setting theory to sport.