Xylose reductase (XR) is a key enzyme in biological xylitol production, and most XRs have broad substrate specificities. During xylitol production from biomass hydrolysate, non-specific XRs can reduce l-arabinose, which is the second-most abundant hemicellulosic sugar, to the undesirable byproduct arabitol, which interferes with xylitol crystallization in downstream processing. To minimize the flux from l-arabinose to arabitol, the l-arabinose-preferring, endogenous XR was replaced by a d-xylose-preferring heterologous XR in Candida tropicalis. Then, Bacillus licheniformis araA and Escherichia coli araB and araD were codon-optimized and expressed functionally in C. tropicalis for the efficient assimilation of l-arabinose. During xylitol fermentation, the control strains BSXDH-3 and KNV converted 9.9 g l-arabinose l(-1) into 9.5 and 8.3 g arabitol l(-1), respectively, whereas the recombinant strain JY consumed 10.5 g l-arabinose l(-1) for cell growth without forming arabitol. Moreover, JY produced xylitol with 42 and 16% higher productivity than BSXDH-3 and KNV, respectively.