One of the greatest avenues opened by next genera&on sequencing (NGS) technologies is the possibility to sequence on large scale species with no prior genomic informa&on, the so called non-model species, at reasonable cost. This has revolu&onized the way biologists approach ques&ons, allowing the implementa&on of experiments previously considered impossible. However, unlike model species, non-model species lack a reference genome or transcriptome assembly, which encouraged the community to develop various strategies to deal with the challenge of construc&ng a reference assembly de novo. Short reads oQen lead to the assembly of incomplete con&gs with low error rate, while longer reads improve greatly the coverage of the sequenced transcripts but are error-prone. A generally recognized need is to increase the length of used reads, but current technologies cannot offer long reads without increasing the error rate at the same &me. Although the combina&on of short and long reads promises to bring together the advantages of both read types, i.e. long reads and low error rate, it is ques&onable whether this is applicable with the available tools. Here, we aim at comparing de novo assemblies incorpora&ng reads produced by different sequencing plaRorms, and in par&cular short reads from Illumina HiSeq2000 (100 bp PE), a bit longer reads produced by Illumina MiSeq (300bp PE), and long reads produced by the PacBio technology (1500 bp). We use as test case the transcriptome sequencing of the whitefly Bemisia tabbaci 1 and assess the efficiency of available soQware developed to combine reads with different lengths. Finally, we provide basic guidelines for improving the assemblies produced for non-model species and propose future direc&ons to deal with this challenge faced by most researchers of organismal biology.