Motivation: Differential network analysis, designed to highlight interaction changes between conditions, is an important paradigm in network biology. However, network analysis methods have been typically designed to compare between few conditions, were rarely applied to protein interaction networks (interactomes). Moreover, large-scale benchmarks for their evaluation have been lacking. Results: Here, we assess five network analysis methods by applying them to 34 human tissues interactomes. For this, we created a manually-curated benchmark of 6,499 tissue-specific, gene ontology biological processes, and analyzed the ability of each method to expose these tissue-process associations. The four differential network analysis methods outperformed the non-differential, expression-based method (AUCs of 0.82-0.9 versus 0.69, respectively). We then created another benchmark, of 1,527 tissue-specific disease cases, and analyzed the ability of differential network analysis methods to highlight additional disease-related genes. Compared to a non-differential subnetworks surrounding a known disease-causing gene, the extremely-differential subnetwork (top 1%) was significantly enriched for additional disease-causing genes in 18.6% of the cases (p<=10E-3). In 5/10 tissues tested, including Muscle, nerve and heart tissues (p = 2.54E-05, 2.71E-04, 3.63E-19), such enrichments were highly significant. Summary: Altogether, our study demonstrates that differential network analysis of human tissue interactomes is a powerful tool for highlighting processes and genes with tissue-selective functionality and clinical impact. Moreover, it offers expansive manually-curated datasets of tissue-selective processes and diseases that could serve for benchmark and for analyses in many other studies.