A complete and high-quality genome reference sequence of an organism provides a solid foundation for a wide research community and determines the outcomes of relevant genomic, genetic, molecular and evolutionary research. Rice is an important food crop and a model plant for grasses, and therefore was the first chosen crop plant for whole genome sequencing. The genome of the japonica representative rice variety, Nipponbare, was sequenced using a gold standard, map-based clone-by-clone strategy. However, although the Nipponbare reference sequence (RefSeq) has the best quality for existing crop genome sequences, it still contains many assembly errors and gaps. To improve the Nipponbare RefSeq, first a robust method is required to detect the hidden assembly errors. Through alignments between BAC-end sequences (BESs) embedded in the Nipponbare bacterial artificial chromosome (BAC) physical map and the Nipponbare RefSeq, we detected locations on the Nipponbare RefSeq that were inversely matched with BESs and could therefore be candidates for spurious inversions of assembly. We performed further analysis of five potential locations and confirmed assembly errors at those locations; four of them, two on chr4 and two on chr11 of the Nipponbare RefSeq (IRGSP build 5), were found to be caused by reverse repetitive sequences flanking the locations. Our approach is effective in detecting spurious inversions in the Nipponbare RefSeq and can be applied for improving the sequence qualities of other genomes as well.