This response is offered to the critique by Gard et al. (2021) of our meta-analysis of polychlorinated biphenyl (PCB)-induced toxicity data in fish (Berninger and Tillitt 2019). Gard et al. (2021) offered numerous comments, the most substantive suggesting that 1) we should have added no-observable–adverse effect residue (NOAER) data from additional studies and all data points from selected studies, and 2) the uncertainty of aggregating data from different PCB mixtures, different species, and different life stages is too great based on a limited data set. The additional studies Gard et al. suggested either were not designed to produce toxicological data, had experimental design issues, were confounded by co-contaminants, or did not contain paired exposure–effects data and as such were not appropriate to add to the data set. Lowest-observable–adverse effect residue (LOAER) values were selected for our analysis because they represent population sensitivities from the central portions of a frequency distribution (the linear portion of dose–response curves). As a consequence, there is less uncertainty in these input data (LOAER values) and greater confidence that they accurately represent the response of fish populations tested. Modeling NOAER values is in the extrapolation portion of a dose–response relationship and subject to enhanced uncertainty. The Gard et al. (2021) critique ignores this fundamental principle of toxicology and adds/deletes data points from our data set without clear selection criteria, which artificially enhances the uncertainty of their models that ultimately are not useful. We reject the premise that it is better to use individual study data as opposed to aggregation of PCB-induced toxicity thresholds in fish.