Understanding the genetic basis of complex traits remains a major challenge in biology. Polygenicity, phenotypic plasticity and epistasis contribute to phenotypic variance in ways that are rarely clear. This uncertainty is problematic for estimating heritability, for predicting individual phenotypes from genomic data, and for parameterizing models of phenotypic evolution. Here we report a recombinant inbred line (RIL) quantitative trait locus (QTL) mapping panel for the hermaphroditic nematode Caenorhabditis elegans, the C. elegans multiparental experimental evolution (CeMEE) panel. The CeMEE panel, comprising 507 RILs, was created by hybridization of 16 wild isolates, experimental evolution at moderate population sizes for 140-190 discrete generations, and inbreeding by selfing for 13-16 generations. The panel contains around 22% of single nucleotide polymorphisms known to segregate in natural populations, and complements existing mapping resources for C. elegans by providing high nucleotide diversity across >95% of the genome, and fine-scale mapping resolution. We apply it to study the genetic basis of two fitness components, fertility and adult hermaphrodite body size, with high broad sense heritability in the CeMEE. While simulations show we should detect common alleles with additive effects as small as 5%, at gene-level resolution, the genetic architectures of these traits does not feature such alleles. We instead find that a significant fraction of trait variance, particularly for fertility, can be explained by sign epistasis with weak main effects. Multiple interactions involving
egl-18 (and potentially the functionally related flanking genes
egl-4 and
elt-6 as well) are highly influential for fertility, consistent with selection for embryo retention during experimental evolution. Phenotype prediction, while generally poor (
r2<10%), requires modeling epistasis for optimal accuracy, with most variance attributed to the highly recombinant, rapidly evolving chromosome arms.