Hypotheses concerning the diversification of notothenioid fishes have relied extensively on estimates of divergence times using molecular clock methods. The timing of diversification of the cold adapted antifreeze glycoprotein (AFGP)-bearing Antarctic notothenioid clade in the middle to late Miocene has been correlated with the onset of polar climatic conditions along the Antarctic Continental Shelf. Critical examination of the previous molecular clock analyses of notothenioids reveals several problems associated with heterogeneity of nucleotide substitution rates among lineages, the application of potentially inappropriate nucleotide substitution rates, and the lack of confidence intervals for divergence time estimates. In this study, the notothenioid partial gene mtDNA 12S-16S rRNA (PG-rRNA) molecular clock was reanalysed using a tree-based maximum likelihood strategy that attempts to account for rate heterogeneity of nucleotide substitution rates among lineages using the penalized likelihood method, and bootstrap resampling to estimate confidence intervals of divergence time estimates. The molecular clock was calibrated using the notothenioid fossil Proeleginops grandeastmanorum. Divergence time estimates for all nodes in the PG-rRNA maximum likelihood tree were substantially older than previous estimates. In particular, the estimated age of the AFGP-bearing Antarctic notothenioid clade predates the onset of extensive sea ice and development of polar conditions by at least 10 million years. Despite caveats involving the fossil calibration and limitations of the PG-rRNA dataset, these divergence time estimates provide initial observations for the development of a novel model of the diversification of cold adapted Antarctic notothenioid fishes.