Due to its economic importance, Coffea arabica is becoming the subject of increasing genomic research and, in particular, the genes involved in the final chemical composition of the bean and the sensorial quality of the coffee beverage. The aim of the present study was to decipher the transcriptional networks that govern the development of the C. arabica seed, a model for non-orthodox albuminous seeds of tropical origin. For this purpose, we developed a transcriptomic approach combining two techniques: targeted cDNA arrays, containing 266 selected candidate gene sequences, and real-time RT-PCR on a large subset of 111 genes. The combination of the two techniques allowed us to limit detection of false positives and to reveal the advantages of using large real-time RT-PCR screening. Multivariate analysis was conducted on both datasets and results were broadly convergent. First, principle component analysis (PCA) revealed a dramatic re-programming of the transcriptional machinery between early cell division and elongation, storage and maturation phases. Second, hierarchical clustering analysis (HCA) led to the identification of 11 distinct patterns of gene expression during seed development as well as to the detection of genes expressed at specific developmental stages that can be used as functional markers of phenological changes. In addition, this study led to the description of gene expression profiles for quality-related genes, most of them formerly uncharacterised in Coffea. Their involvement in storage compound synthesis and accumulation during endosperm development and final metabolic re-adjustments during maturation is discussed.