This study's hypotheses were that both shortly after instruction and after an 8-month follow-up, diabetic children taught via anchored instruction (AI), a format for problem solving, would outperform controls. Subjects were 81 9-15-year-old campers with insulin dependent diabetes who were randomly assigned to AI or control groups for two 45-min small group teaching sessions. Als viewed a video about a girl who mismanages her diabetes during intercurrent illness, and they were challenged to identify, define and solve her errors. Controls learned sick-day management via conventional direct instruction. At the end of the 2-week camp, AI and control groups' scores on factual knowledge were equal. AIs were more likely than controls at the end of the camp (0.75 vs. 0.54, P < 0.05) and 8 months later (0.59 vs. 0.38, P < 0.02) to provide a rationale for the use of remembered guidelines. Across all campers, this ability to link guidelines and their rationales was significantly correlated (r = 0.55, P < 0.01) with the number of self-management practices employed by campers who suffered an illness between the end of camp and the 8-month follow-up. Only one long-term behavioral difference between groups emerged: Als' parents shared in making most diabetes decisions on sick days, while controls' parents left more decision making to their children. AI appears at least as good as conventional teaching, and may better 'link' rules and reasons, perhaps aiding daily real-life problem solving.