By a computational approach we reconstructed genomic transcriptional profiles of 19 different adult human tissues, based on information on activity of 27,924 genes obtained from unbiased UniGene cDNA libraries. In each considered tissue, a small number of genes resulted highly expressed or "tissue-specific". Distribution of gene expression levels in a tissue appears to follow a power law, thus suggesting a correspondence between transcriptional profile and "scale-free" topology of protein networks. The expression of 737 genes involved in Mendelian diseases was analysed, compared with a large reference set of known human genes. Disease-genes resulted significantly more expressed than expected. The possible correspondence of their products to important nodes of intracellular protein network is suggested. Auto-organization of the protein network, its stability in time in the differentiated state and relationships with the degree of genetic variability at genome level are discussed.