Association studies between genetic variants and complex traits are popular and valuable in both genetic and clinical fields. Among all kinds of studies proposed, transcriptome-wide association studies (TWAS) have become influential and widely used. In my thesis, I focus on revealing under which settings of genetic parameters and architectures, TWAS will be more powerful in detecting contributing genes than other analytical methods, including genome-wide association studies (GWAS) and eQTL-based meditated GWAS (emGWAS). We first derive novelly the closed-form of the non-centrality parameter (NCP) in the non- central distribution under alternative hypothesis. Then we estimate the power based on the estimated NCP. Through simulation studies, we compare the power of the three methods, i.e. TWAS, GWAS and emGWAS. Our numerical results show that while the number of significant genes, level of trait heritability and phenotypic variance component explained by expressions (PVX) all have influence on the power of the three analytical models according to the corresponding genetic architecture, the expression heritability is the most influential factor which makes TWAS stand out.