Selection coefficients are estimates of the magnitude and direction of selection through patterns of allele frequency change over time. There exist numerous methods to estimate selection coefficients from time-series genomic data. However, these tools can be limited in their ability to accurately infer selection if the underlying model assumptions are not met by populations. Most tools implement the assumptions of a Wright-Fisher (WF) model in order to increase computational efficiency. A basic WF model operates under the following assumptions: (1) non-overlapping generations, (2) constant population size, (3) random mating, (4) no mutation, and (5) no migration. The WF model has provided the foundation for describing many theories in population genetics and evolutionary biology, but few, if any, populations adhere to even just some of its assumptions. Evaluations of tools thus far have focused almost solely on cases where population size is assumed to remain constant. Evolutionary rescue is an ecologically relevant scenario in which populations experience fluctuating demography. Evolutionary rescue typically results from maladaptation to an environmental change where the mean fitness of the population is << 1, leading to a decline in abundance. Subsequently, through adaptive evolution, the population increases in fitness to such a degree that the population is rescued from extinction. Cases of successful evolutionary rescue imply that the rescued population experienced rapid changes in census and effective population size, breaking the assumptions of typical WF models. Despite strong selection, the effects of genetic drift may come to dominate evolutionary dynamics during severe bottlenecks. Bottlenecks and strong selection can also cause extreme shifts in linkage disequilibria (LD) that affect the dynamics of linked sites. Recent work has also shown that evolutionary rescue events can increase the rates of selective sweeps and harden sweeps, which could have significant consequences to methods dependent on characterizing selection through time-series patterns. Here, we assess the accuracy of one popular method, WFABC, in which we estimate selection coefficients under the ecologically relevant scenario of evolutionary rescue. We measure the accuracy of this method based on simulated genomic data in which the selection coefficients used to generate allele frequencies are known. Our study first estimates selection coefficients in single-site simulations, representing an idealized case in which the effects of epistasis or physical linkage do not influence the estimate of selection. We then implement a multi-site model in which multiple selected loci present at once may influence the dynamics of surrounding sites through linkage while still excluding epistasis.