Single genetic variation may only have a modest effect on risk of gastric cardia adenocarcinoma (GCA) because this malignancy is believed to result from complex interactions among multiple genetic and environmental factors. However, it has been a challenge to characterize multiple interactions using parametric analytic approaches. This study utilized a multi-analytic strategy combining logistic regression (LR), multifactor dimensionality reduction (MDR) and classification and regression tree (CART) approaches to explore high-order interactions among smoking and 12 polymorphisms involved in different processes of carcinogenesis in 344 GCA patients and 324 controls. LR, MDR and CART analyses consistently suggested MMP-2 C-1306T polymorphism as the strongest individual factor for GCA risk. Intriguingly, a high-order interaction was consistently identified by MDR, LR and CART analyses. In MDR analysis, the three-factor model including MMP-2 C-1306T, FASL T-844C and FAS G-1377A yielded the highest testing accuracy of 0.632. When analysing combined effect of these three polymorphisms by LR, a significant gene dose effect was observed with the odds ratios (ORs) being increased with increasing numbers of risk genotypes (P(trend) = 4.736 × 10⁻¹²). In CART analysis, individuals carrying the combined genotypes of MMP-2 -1306CC, FASL-844TT or TC and FAS -1377AA had the highest risk for GCA (OR = 4.58; 95% confidence interval, 2.07-10.14) compared with the lowest risk carriers of the MMP-2 -1306CT or TT genotype. These results suggest that MMP-2 C-1306T polymorphism is an important risk factor for GCA and the multifactor interactions among polymorphisms in MMP-2, FASL and FAS play more important role in the development of GCA.