Type 2 diabetes (T2D) is a major global health burden, and dietary modification is a cornerstone of its prevention. To strengthen causal inference in nutrition research, we recently developed a Mendelian randomization (MR) framework based on chemosensory genetics to identify biologically meaningful genetic instruments for dietary exposures. Using this approach, we identified a genetic instrument for onion liking that was independently validated through its association with onion intake and showed that genetically predicted onion liking is associated with lower blood pressure (BP) and a lower risk of T2D(1). Given that elevated BP is an established risk factor for T2D, we hypothesised that BP mediates the protective association between onion consumption and T2D. Here, we investigated whether the association between genetically predicted onion liking and T2D is mediated through BP and independent of glycaemic traits. We also evaluated whether the association could be explained by genetically correlated food preferences. Two-sample MR and mediation analyses were performed using large-scale genome-wide association study summary statistics for onion liking (n = 159,518), blood pressure (n = 757,601), glycaemic traits (n = 200,622) and T2D (n = 242,283 cases; 1,569,734 controls). Multivariable MR (MVMR) was performed to evaluate whether the observed associations were independent of glycaemic traits and genetically correlated vegetable preferences. Consistent with our previous findings, genetically predicted onion liking was associated with a lower risk of T2D (OR = 0.86, 95% CI 0.78–0.94) and lower systolic BP (SBP) and diastolic BP (DBP), but not fasting glucose, fasting insulin or HbA1c levels. Mediation analyses indicated that SBP partially mediated the association between onion liking and T2D (29.2%, 95% CI 6.3–52.0%), but not through DBP (8.3%, 95% CI -2.5–19.1%). MVMR confirmed that the causal effect of SBP on T2D remained after adjustment for fasting glucose, fasting insulin and HbA1c, suggesting that SBP is likely acting through a non-glycaemic pathway. Moreover, the association between onion liking and T2D was attenuated after conditioning on SBP. Reverse MR found no evidence that genetically predicted SBP influenced onion liking, supporting the inferred direction of the onion–SBP association. MVMR analyses accounting for genetically correlated food preferences yielded directionally consistent estimates with overlapping confidence intervals, suggesting that the association was not fully explained by shared genetic influences on other food preferences. These findings suggest that the protective association between onion consumption and T2D may be mediated, in part, through BP regulation rather than conventional glycaemic pathways. More broadly, this study illustrates how biologically informed MR can strengthen causal inference and help elucidate causal pathways linking individual foods with chronic disease.