Background: Recent advances in genomics indicate functional significance of a majority of genome sequences and their long range interactions. As a detailed examination of genome organization and function requires very high quality genome sequence, the objective of this study was to improve reference genome assembly of banana (Musa acuminata). Results: We have developed a modular bioinformatics pipeline to improve genome sequence assemblies, which can handle various types of data. The pipeline comprises several semi-automated tools. However, unlike classical automated tools that are based on global parameters, the semi-automated tools proposed an expert mode for a user who can decide on suggested improvements through local compromises. The pipeline was used to improve the draft genome sequence of Musa acuminata. Genotyping by sequencing (GBS) of a segregating population and paired-end sequencing were used to detect and correct scaffold misassemblies. Long insert size paired-end reads identified scaffold junctions and fusions missed by automated assembly methods. GBS markers were used to anchor scaffolds to pseudo-molecules with a new bioinformatics approach that avoids the tedious step of marker ordering during genetic map construction. Furthermore, a genome map was constructed and used to assemble scaffolds into super scaffolds. Finally, a consensus gene annotation was projected on the new assembly from two pre-existing annotations. This approach reduced the total Musa scaffold number from 7513 to 1532 (i.e. by 80 %), with an N50 that increased from 1.3 Mb (65 scaffolds) to 3.0 Mb (26 scaffolds). 89.5 % of the assembly was anchored to the 11 Musa chromosomes compared to the previous 70 %. Unknown sites (N) were reduced from 17.3 to 10.0 %. Conclusion: The release of the Musa acuminata reference genome version 2 provides a platform for detailed analysis of banana genome variation, function and evolution. Bioinformatics tools developed in this work can be used to impr