Fusion transcripts that are frequent in cancer can be exploited to understand the mechanisms of malignancy and can serve as diagnostic or prognostic markers. Several algorithms have been developed to predict fusion transcripts from DNA or RNA data. The majority of these algorithms align sequencing reads to the reference transcriptome for predicting fusions; however, this results in several undetected fusions due to the highly perturbed nature of cancer genomes. Here, we describe a novel method that uses a k-mer based algorithm to predict fusion transcripts accurately using the unaligned reads from the regular RNA-seq data analysis pipelines.