Background: The quality of gene annotation determines the interpretation of results obtained in transcriptomic studies. The growing number of genome sequence information calls for experimental and computational pipelines for de novo transcriptome annotation. Ideally, gene and transcript models should be called from a limited set of key experimental data. Results: We developed TranscriptomeReconstructoR, an R package which implements a pipeline for automated transcriptome annotation. It relies on integrating features from independent and complementary datasets: (i) full-length RNA-seq for detection of splicing patterns and (ii) high-throughput 5′ and 3′ tag sequencing data for accurate definition of gene borders. The pipeline can also take a nascent RNA-seq dataset to supplement the called gene model with transient transcripts. We reconstructed de novo the transcriptional landscape of wild type Arabidopsis thaliana seedlings and Saccharomyces cerevisiae cells as a proof-of-principle. A comparison to the existing transcriptome annotations revealed that our gene model is more accurate and comprehensive than the most commonly used community gene models, TAIR10 and Araport11 for A.thaliana and SacCer3 for S.cerevisiae. In particular, we identify multiple transient transcripts missing from the existing annotations. Our new annotations promise to improve the quality of A.thaliana and S.cerevisiae genome research. Conclusions: Our proof-of-concept data suggest a cost-efficient strategy for rapid and accurate annotation of complex eukaryotic transcriptomes. We combine the choice of library preparation methods and sequencing platforms with the dedicated computational pipeline implemented in the TranscriptomeReconstructoR package. The pipeline only requires prior knowledge on the reference genomic DNA sequence, but not the transcriptome. The package seamlessly integrates with Bioconductor packages for downstream analysis.