Studies investigating the genetics of gene expression, or expression quantitative trait loci (eQTLs), have aided the interpretation of human disease-associated genetic variants, many of which lie in non-coding regions. Although a great number of eQTLs have been identified, many key aspects of the study design and analysis, such as the rate of false discoveries, have not been well studied. We used extensive simulations of matched gene expression and human genetic variation to investigate how sample size and allele frequency affect power and effect size estimation. Genotype data was simulated using HAPGEN2 at varying sample sizes together with a 1000 Genomes reference of chromosome 22 to mimic realistic linkage disequilibrium (LD) patterns. In each scenario, true cis eQTLs were randomly sampled from SNPs with a fixed minor allele frequency (MAF). Expression data were simulated using a simple linear model, where the genetic effect sizes were drawn from a distribution obtained from a real dataset. As expected, power increased as eQTL MAF, effect size, or the study sample size increased. In comparing the performance of various multiple testing correction methods, we observed that local LD structure around cis eQTLs inflated the false discovery rate (FDR), sometimes substantially. Implementation of on hierarchical testing procedures recalibrated the FDR with minimal effect power. We further observed the frequent overestimation of eQTL effect size, the so-called “Winner’s Curse”. Winner’s Curse was more severe when the statistical power of eQTL detection was low. To address this, we compared various estimators using two resampling-based methods and found that an adapted bootstrap weighted estimator performed best in terms of reduced upward effect size bias and smaller variance. The insights from our study are likely useful for future eQTL study design and analysis strategy.