Drought stress is a significant production constraint for rainfed agriculture system. Bananas grown in the tropics and sub-tropics without irrigation suffer serious yield reductions attributable to moisture stress. Although a reference genome for banana would be useful, a reference transcriptome can assist our understanding of banana response to water deficit at the molecular level. In this study we present a reference transcriptome generated from cDNA libraries from root and leaf samples from Mbwazirume ‘AAA’ and Cachaco ‘ABB’ maintained at pF 1.8-2.1 (well-watered) and pF 2.8-3.1 (dry). Additional samples were taken from flowers and fruits from the same genotypes grown in the field. Sequencing resulted in about 600,000 reads for each cultivar, giving a total sequence length of about 375 Mbp. The combined assembly contained 25,250 large contigs ranging from <500 to 4000 bp, based on 3 to 900 reads per contig. The majority of large contigs contained intact open reading frames (iORF). BLASTP (www.ebi.ac.uk/Tools/sss/ncbiblast/) revealed higher similarity of Musa gene products to those of Oryza sativa, Hordium vulgare (barley), and Sorghum bicolor than for other monocots like Zea mays. This reference transcriptome will be useful for (i) analysis of differential gene expression studies among bananas, (ii) discovery of SNPs and INDELs that could be used as markers in other studies, and (iii) discovery of alternative splicing in expressed genes. We expect to improve this reference transcriptome with Illumina sequences from the same genotypes under similar conditions and characterize the metabolic pathways affected by moisture stress.