Objective To systematically analyze the correlation between smoking and Alzheimer's disease (AD) based on different study designs of case-control study and prevalence study and evaluate the design methods. Methods English literature was retrieved from Cochrane Library and Medline and Chinese literature was retrieved from CBMDISC. Cases that met the NINCDS-ADRDA diagnose criteria were collected. Case control study, prevalence study, and meta-analysis were conducted. Results A total of 25 individual studies were selected, including 2 matched case-control studies, 21 parallel case-control studies, and 2 cross-sectional studies. Among all studies, 5 case-control studies only reported crude odds ratios, 16 case-control studies reported crude and adjusted odds ratios, and 25 studies yielded a pooled odds ratio of 0.877 (95%CI: 0.794, 0.969). After eliminating studies with crude odds ratios, 20 studies, which included 2 matched case-control studies, 16 parallel case control studies, and 2 cross-sectional studies, still yielded a pooled odds ratio of 0.886 (95%CI: 0.798, 0.985). After eliminating 2 cross-sectional studies, 18 case-control studies yielded a pooled odds ratio of 0.910 (95%CI: 0.816, 1.016) and was not statistically significant, while 2 cross-sectional studies yielded a pooled odds ratio of 0.641 (95%CI: 0.436, 0.941), which was statistically significant. Sixteen parallel case-control studies and 2 matched case-control studies yielded a pooled odds ratio of 0.911 (95%CI: 0.814, 1.019) and 0.896 (95%CI: 0.547, 1.466), respectively. Sixteen case-control studies yielded a pooled odds ratio of 0.739 (95%CI: 0.584, 0.934) for crude odds ratio and 0.911 (95%CI: 0.814, 1.019) for adjusted odds ratio, respectively, and the difference was statistically significant. A meta regression showed that the differences of pooled odds ratio of parallel case-control studies, matched case-control studies, and prevalence design studies were statistically significant. Begg's funnel plot and Egger's test showed that there was no evidence of publication bias. Conclusion Matched case-control studies and parallel case-control studies show that smoking is not an influencing factor of the incidence of AD. Crude odds ratio and crosssectional study design decrease the odds ratio of correlation between smoking and AD, but can not be regarded as an evidence of causal relationship.