We applied a simple and efficient two-step method to analyze a family-based association study of gene expression quantitative trait loci (eQTL) in a mixed model framework. This two-step method produces very similar results to the full mixed model method, with our method being significantly faster than the full model. Using the Genetic Analysis Workshop 15 (GAW15) Problem 1 data, we demonstrated the value of data filtering for reducing the number of tests and controlling the number of false positives. Specifically, we showed that removing non-expressed genes by filtering on expression variability effectively reduced the number of tests by nearly 50%. Furthermore, we demonstrated that filtering on genotype counts substantially reduced spurious detection. Finally, we restricted our analysis to the markers and transcripts that were closely located. We found five times more signals in close proximity (cis-) to transcripts than in our genome-wide analysis. Our results suggest that careful pre-filtering and partitioning of data are crucial for controlling false positives and allowing detection of genuine effects in genetic analysis of gene expression.