Recently introduced, linked-read technologies, such as the 10X chromium system, use microfluidics to tag multiple short reads from the same long fragment (50-200 kbp) with a small sequence, called a barcode. They are inexpensive and easy to prepare, combining the accuracy of short-read sequencing with the long-range information of barcodes. The same barcode can be used for several different fragments, which complicates the analyses. We present QuickDeconvolution (QD), a new software for deconvolving a set of reads sharing a barcode, i.e. separating the reads from the different fragments. QD only takes sequencing data as input, without the need for a reference genome. We show that QuickDeconvolution outperforms existing software in terms of accuracy, speed, and scalability, making it capable of deconvolving previously inaccessible data sets. In particular, we demonstrate here the first example in the literature of a successfully deconvoluted animal sequencing dataset, a 33 Gbp Drosophila melanogaster dataset. We show that taxonomic assignment of linked reads can be improved by deconvoluting reads with QD before taxonomic classification.