Knowledge of recombination rate variation along the genome provides important insights into genome and phenotypic evolution. Population genomic approaches offer an attractive way to infer the population-scaled recombination rate ρ =4 N e r using the linkage disequilibrium information contained in DNA sequence polymorphism data. Such methods have been used on a broad range of plant and animal species to build genome-wide recombination maps. However, the reliability of these inferences has only been assessed under a restrictive set of conditions. Here, we evaluated the ability of one of the most widely used coalescent-based approaches, LDhelmet , to infer a biologically-realistic genomic landscape of recombination. Using simulations, we specifically assessed the impact of empirical (sample size, block penalty) and evolutionary parameters (effective population size ( N e ), mutation and recombination rates) on inferred map quality. We report reasonably good correlations between simulated and inferred landscapes, but point to limitations when it comes to detecting recombination hotspots. False positives and false negatives considerably confound fine-scale patterns of inferred recombination under a wide array of conditions, particularly when N e is small and the mutation/recombination rate ratio is low, to the extent that maps inferred from populations sharing the same recombination landscape appear uncorrelated. We thus address a message of caution to users of such approaches, while also recognizing their importance and potential, particularly in species with less complex landscapes for which LD-based approaches should provide high quality recombination maps.