Detecting structural variants (SVs) from sequencing data is key to genome analysis, but methods using standard whole-genome sequencing (WGS) data are typically incapable of resolving complex SVs with multiple co-located breakpoints. We introduce the ARC-SV method, which uses a probabilistic model to detect arbitrary local rearrangements from WGS data. Our method performs well on simple SVs while surpassing state-of-the-art methods in complex SV detection.Since observations of microscopically visible aneuploidies and gene duplications, the study of large-scale genomic alterations has been a key component of genome analysis [1]. Structural variants, typically defined as mutations affecting at least 50 bp of sequence, are more rare than single-nucleotide variants (SNVs), but SVs are known to account for a much larger portion of sequence difference between human individuals [2,3]. Functionally, SVs are also more likely than SNVs to impact gene expression, with larger effect sizes on average [4,5]. The rapid improvements to sequencing technologies have prompted development of numerous SV detection methodologies [6], some of which have been applied in population-scale sequencing projects [4,7].The vast majority of sequencing-based SV callers detect deletions, duplications, inversions, and/or translocations [6]. If only these SV types are present, any SV can be identified by a single breakpoint, or pair of bases adjacent in the sample but not the reference. Detection of these breakpoints is theoretically straightforward given high coverage data and accurate alignments. Misaligned reads make SV detection difficult in practice, and investigators typically apply multiple methods together with heuristic filters in order to achieve high accuracy [4,[7][8][9][10]. Structural variants with complexity beyond the scope of most detection algorithms have also been observed in a variety of phenotypic contexts [4,[11][12][13]. Our definition of a complex SV (cxSV) is any rearrangement not reducible to non-overlapping deletions, tandem duplications, novel insertions, and inversions; we focus here on the localized cxSVs commonly observed in the germline.