Problems associated with medication use and the consequent effects on genome-wide association analyses were explored using the Genetic Analysis Workshop 16 Problem 3 data. Lipid phenotypes were simulated in the Framingham Heart Study using several measured variables including causal genes (based on a 500 k SNP panel), smoking, dietary intake, and medication usage. We report a sensitivity analysis of how medication use (which artificially alters lipid values) affects association inferences. Associations were performed for LDL-c under seven different correction schemes: 1) ignore medication use entirely (no correction) and adjust for covariates; 2) delete medicated subjects then adjust for covariates; 3) include medication use (Yes/No) as a covariate during covariate adjustments; 4) correct raw values using clinical trials information then adjust for covariates; 5) correct raw values using the actual simulation protocol ("truth") then adjust for covariates; and 6-7) over-corrections (add arbitrary values) then adjust for covariates. Results indicate that failure to properly correct for medication usage can profoundly affect the heritability, and hence the association results. The empirical results yielded one genome-wide significant locus on chromosome 22 (RS2294207), consistent with the simulation protocol. This signal was detected under all schemes that corrected the raw values (clinical trials, simulation protocol, or over corrections), but was not detected under the first three adjustment schemes (ignore medication use, delete medicated individuals, use medication status as covariate). In summary, we confirm that failure to properly account for medication usage can have a profound impact on genetic associations.