It is believed that large numbers of genes are involved in common human brain diseases. Here, we propose a novel computational strategy for simultaneously identifying multiple candidate genes for genetic human brain diseases from a brain-specific gene network-level perspective. By integrating diverse genomic and proteomic datasets based on Bayesian statistical model, we built a large-scale human brain-specific gene network. Based on this network and minor prior knowledge of a specific brain disease, we can effectively identify multiple candidate genes for this disease. When four known Alzheimer's disease genes were used as the prior knowledge, among the top 46 high-scoring genes that we have found, 37 were previously reported to be associated with Alzheimer's disease. And the higher score a gene has, the more likely this gene is a disease-related one. The results suggest that the proposed method is effective, convenient, and applicable in the future genetic studies.