This study aims to assess the effect of familial structures on the still-missing heritability estimate and prediction accuracy of Type 2 Diabetes (T2D) using pedigree estimated risk values (ERV) and genomic ERV. We used 11,818 individuals (T2D cases: 2,210) with genotype (649,932 SNPs) and pedigree information from the ongoing periodic cohort study of the Iranian population project. We considered three different familial structure scenarios, including i) all families, ii) all families with ≥ 1 generation, and iii) families with ≥ 1 generation in which both case and control individuals are presented. Comprehensive simulation strategies were implemented to quantify the difference between estimates of h2 and h2 NSP). A proportion of still-missing heritability in T2D could be explained by overestimation of pedigree-based heritability due to the presence of families with individuals having only one of the two disease statuses. Our results indicated noticeable differences in prediction accuracy for different family structures, where families with generation ≥ 1 and having both cases and controls showed the highest prediction accuracy and the highest correlation with Polygenic Risk Scores. Our findings represent the first evidence of the important contribution of familial structure for heritability estimations and genomic prediction studies in T2D.