Simultaneous exposure to a mixture of chemicals over a lifetime increases an individual's risk of disease. Weighted quantile sum (WQS) regression estimates the effect of exposure to chemicals and identifies the important chemicals in the mixture. However, complications arise when experimental apparatus can only detect chemical concentrations to a detection limit. In WQS analyses, these values below the detection limit (BDLs) are either placed in the first quantile of the weighted index (BDLQ1) or imputed by bootstrapping or Bayesian imputation. However, the impact of these approaches on inference is unknown. We compared these approaches via a simulation study over 0% (baseline), 10%, 33%, 50%, and 80% BDL for each chemical. The true mixture consisted of 14 chemicals with 4 important chemicals. We examined the ability of each method to estimate the mixture effect and to identify the chemicals most strongly associated with disease. The imputation methods performed as well as or better than BDLQ1 in sensitivity and specificity. They also had lower MSE for the health effect until 80% BDL.