Fracture-classification systems are used to recommend treatment and predict outcomes. In this study, a modified Gartland classification system of supracondylar humerus fractures in children was assessed for intraobserver and interobserver variability. Five observers classified radiographs of 50 consecutive children with extension supracondylar humerus fractures on three separate occasions. After a 2-week interval, 90% of fractures were classified the same on both readings, with and intraobserver kappa value of 0.84. After a 36-week interval, 89% of the fractures were classified the same, with a kappa value of 0.81. Interobserver reliability was evaluated by pairwise comparison among observers, resulting in an overall kappa value of 0.74. The reliability of the Gartland classification for supracondylar humerus fractures in children is better than that published for other fracture-classification systems. However, 10% of the time, a second reading by the same observer is different. This makes treatment recommendations based only on fracture type imprecise.