The widespread use of High-Throughput Sequencing (HTS) for detection of plant viruses and sequencing of plant virus genomes has led to the generation of large amounts of data and of bioinformatics challenges to process them. Many bioinformatics pipelines for virus detection are available, making the choice of a suitable one difficult. A robust benchmarking is needed for the unbiased comparison of the pipelines, but there is currently a lack of reference datasets that could be used for this purpose. We present 7 semi-artificial datasets composed of real RNA-seq datasets from virus-infected plants spiked with artificial virus reads. Each dataset addresses challenges that could prevent virus detection. We also present 3 real datasets showing a challenging virus composition as well as 8 completely artificial datasets to test haplotype reconstruction software. With these datasets that address several diagnostic challenges, we hope to encourage virologists, diagnosticians and bioinformaticians to evaluate and benchmark their pipeline(s).