There is a long-standing interest in understanding host-parasite coevolutionary dynamics and associated fitness effects. Increasing amounts of genomic data for both interacting species offer a promising source to identify candidate loci and to infer the main parameters of the past coevolutionary history. However, so far no method exists to do so. By coupling a gene-for-gene model with coalescent simulations, we first show that three types of biological cost, resistance, infectivity and infection, define the allele frequencies at the internal equilibrium point of the coevolution model, which in return determine the strength of the selective signatures signatures at host and parasite loci. We apply an Approximate Bayesian Computation (ABC) approach on simulated datasets to infer these costs by jointly integrating host and parasite polymorphism data at the coevolving loci. To control for the effect of genetic drift on coevolutionary dynamics, we assume that 10 or 30 repetitions are available from controlled experiments or several natural populations. We study two scenarios: 1) the cost of infection and population sizes (host and parasite) are unknown while costs of infectivity and resistance are known, and 2) all three costs are unknown while populations sizes are known. Using the ABC model choice procedure, we show that for both scenarios, we can distinguish with high accuracy pairs of loci from host and parasite under coevolution from neutrally evolving loci, though the statistical power decreases with higher cost of infection. The accuracy of parameter inference is also very high under both scenarios especially when using both host and parasite data because parasite polymorphism data do inform on host costs and vice-versa. As the false positive rate to detect genes under coevolution is small, we suggest to use our method to identify host and parasite candidate loci for further functional studies.Author summary It is of importance for agriculture and medicine to understand host-parasite antagonistic coevolutionary dynamics and the deleterious associated fitness effects, as well as to reveal the genes underpinning these interactions. The increasing amount of genomic data for hosts and parasites offer a promising source to identify such candidate loci, but also to use statistical inference methods to reconstruct the past coevolutionary history. In our study we attempt to draw inference of the past coevolutionary history at key host and parasites loci using sequence data from several individuals and across several experimental replicates. We demonstrate that using a Bayesian statistical method, it is possible to estimate the parameters driving the interaction of hosts and parasites at these loci for thousands of generations. The main parameter that can be estimated is the fitness loss by hosts upon infection. Our method and results can be applied to experimental coevolution data with sequences at the key candidate loci providing enough repetitions and large enough population sizes. As a proof of principle, our results open the door to reconstruct past coevolutionary dynamics using sequence data of interacting species.