Sequencing of RNA provides the possibility to study an individual’s transcriptome landscape and determine allelic expression ratios. Single-molecule protocols generate multi-kilobase reads longer than most transcripts allowing sequencing of complete haplotype isoforms. This allows partitioning the reads into two parental haplotypes. While the read length of the single-molecule protocols is long, the relatively high error rate limits the ability to accurately detect the genetic variants and assemble them into the haplotype-specific isoforms. In this paper, we present HapIso (Haplotype-specific Isoform Reconstruction), a method able to tolerate the relatively high error-rate of the single-molecule platform and partition the isoform reads into the parental alleles. Phasing the reads according to the allele of origin allows our method to efficiently distinguish between the read errors and the true biological mutations. HapIso uses a k-means clustering algorithm aiming to group the reads into two meaningful clusters maximizing the similarity of the reads within cluster and minimizing the similarity of the reads from different clusters. Each cluster corresponds to a parental haplotype. We use family pedigree information to evaluate our approach. Experimental validation suggests that HapIso is able to tolerate the relatively high error-rate and accurately partition the reads into the parental alleles of the isoform transcripts. Furthermore, our method is the first method able to reconstruct the haplotype-specific isoforms from long single-molecule reads. The open source Python implementation of HapIso is freely available for download at https://?github.?com/?smangul1/?HapIso/?.