Background: High-throughput RNA sequencing techniques allow the identification of long non-coding RNAs (lncRNAs) that respond to different conditions and have biomarker potential. In non-model organisms, such as commercial sugarcane cultivars, the absence of a sequenced genome makes it difficult to perform analyses based solely on mapping. In contrast, de novo assembly of contigs from reads can generate false positives. To overcome these limitations, we combined these two approaches: (i) de novo assembly of contigs from reads of four cultivars with Trinity Assembly and (ii) mapped the transcriptomic reads to the available genome of cultivar R570 using HISAT2+StringTie. The pipeline for lncRNA prediction was also adjusted for each approach. Considering the total set of lncRNAs across the four cultivars, de novo assembly produced 1,043,229 contigs, while mapping produced 28,833. In addition, BLASTn between these sets indicates that 88,871 contigs (8,51%) from the de novo assembly matched the mapped set (qcovs ≥ 80%). The combination of approaches captures transcripts conserved with the R570 genome and brings robustness to the predicted lncRNAs in the sugarcane transcriptome.