目標：使用多層次分析探討學校菸害防制政策與個人層次特質因素對於學童吸菸行為之相關性。方法：採多步驟集束抽樣選取屏東縣26間國小，以學校為單位，於2008年調查國小三至六年級學童之吸菸行為及其相關影響因素、並同時收集學校層次結構性與菸害防制政策特質因素。共收集有效學生樣本數2350、家長樣本數3726與教職員樣本數52位。結果：學校層次特質及學童本身特質，可解釋學童吸菸行為變異量分別為85%與48%。多層次迴歸模式發現學校層次因素，含山地地區學校(AOR＝2.40, 95%CI: 1.00-5.79)、有教職員吸菸的學校(AOR＝2.57, 95%CI: 1.59-4.17)及未推廣戒菸輔導或活動的學校(AOR＝2.88, 95%CI:1.39-5.94)對於學童個人吸菸行為具有顯著影響。同時自覺他人吸菸頻率、朋友與家人在面前吸菸頻率、喝酒與嚼食檳榔之行為等個人層次變項，亦為顯著影響因素。結論：學校層次特質與國小學童吸菸行為顯著相關，建議教育與衛生機構能重視這些高風險學校及落實校園菸害防制政策與推廣活動，避免國小學童吸菸行為之發生。

Objectives: The aim of this study was to use a multilevel analysis to assess the school-level tobacco control policy and individual-level characteristics of students in relation to cigarette smoking among elementary schoolchildren. Methods: We conducted a multilevel study to assess two-level effects on smoking among 2,350 grade 3-6 students in 26 randomly selected elementary schools in southern Taiwan. The individual- and school-level characteristics regarding smoking were estimated using multi-level logistic regression. Results: School- and individual-level effects, accounting for 85% and 48% respectively of variance in the odds of being a smoker, indicated that the school cluster was very important. The risk of smoking was significantly associated with those schools located in a mountainous region [adjusted odds ratio (AOR=2.40, 95%CI:1.00-5.79)], schools where staff smoked (OR=2.57, 95%CI:1.59-4.17) and schools without anti-tobacco activities or curricula implementation (AOR=2.88, 95%CI:1.39-5.94). Individual-level characteristics having a significant relationship with student smoking behavior were student’s perceptions of smoking, having best friends or family always smoking in front of them, ever-drinking and -chewing betelnut behavior. Conclusions: The findings suggest that tobacco control policies and preventive programs should be strongly considered for implementation in those elementary schools where students are at the greatest risk for cigarette smoking.