Empirical population genetic studies generally rely on sampling subsets of the population(s) of interest and of the nuclear or organellar genome targeted, assuming each are representative of the whole. Violations of these assumptions may impact population-level parameter estimation and lead to spurious inferences. Here we used targeted capture to sequence the full mitochondrial genome from 123 individuals of the Galapagos giant tortoise endemic to Pinzón Island (Chelonoidis duncanensis) sampled at two time points pre- and post-bottleneck (circa 1906 and 2014) to explicitly assess differences in diversity estimates and demographic reconstructions based on subsets of the mitochondrial genome versus the full sequences, and to evaluate potential biases associated with diversity estimates and demographic reconstructions from post-bottlenecked samples alone. Haplotypic diversities were equal between the temporal samples based on the full mitochondrial genome, but single gene estimates suggested either decreases or increases in diversity depending upon the region. Demographic reconstructions based on the full sequence were more similar between the temporal samples than those based on the control region alone, or a subset of three regions, where the trends in population size changes shifted in magnitude and direction between the temporal samples. In all cases, the estimated coalescent point was more distant for the historical than contemporary sample. In summary, our results empirically demonstrate the influence of sampling bias when interpreting population genetic patterns and punctuate the need for careful consideration of potentially conflicting evolutionary signal across the mitochondrial genome.