Abstract Motivation Diploid assembly, or determining sequences of homologous chromosomes separately, is essential to elucidate genetic differences between haplotypes. One approach is to call and phase single nucleotide variants (SNVs) on a reference sequence. However, this approach becomes unstable on large segmental duplications (SDs) or structural variations (SVs) because the alignments of reads deriving from these regions tend to be unreliable. Another approach is to use highly accurate PacBio HiFi reads to output diploid assembly directly. Nonetheless, HiFi reads cannot phase homozygous regions longer than their length and require oxford nanopore technology (ONT) reads or Hi-C to produce a fully phased assembly. Is a single long-read sequencing technology sufficient to create an accurate diploid assembly? Results Here, we present JTK, a megabase-scale diploid genome assembler. It first randomly samples kilobase-scale sequences (called ‘chunks’) from the long reads, phases variants found on them, and produces two haplotypes. The novel idea of JTK is to utilize chunks to capture SNVs and SVs simultaneously. From 60-fold ONT reads on the HG002 and a Japanese sample, it fully assembled two haplotypes with approximately 99.9% accuracy on the histocompatibility complex (MHC) and the leukocyte receptor complex (LRC) regions, which was impossible by the reference-based approach. In addition, in the LRC region on a Japanese sample, JTK output an assembly of better contiguity than those built from high-coverage HiFi+Hi-C. In the coming age of pan-genomics, JTK would complement the reference-based phasing method to assemble the difficult-to-assemble but medically important regions. Availability and implementation JTK is available at https://github.com/ban-m/jtk, and the datasets are available at https://doi.org/10.5281/zenodo.7790310 or JGAS000580 in DDBJ.