DNA copy number alterations (CNA), which are amplifications and deletions of certain regions in the genome, play an important role in the pathogenesis of cancer and have been shown to be associated with other diseases such as autism, schizophrenia and obesity. Next-generation sequencing technologies provide an opportunity to identify CNA regions with unprecedented accuracy. We developed a CNA detection algorithm based on single-end whole-genome sequencing data for samples with matched controls. This algorithm, called BIC-seq, can accurately and efficiently identify the CNAs via minimizing the Bayesian information criterion (BIC). We applied BIC-seq on a glioblastoma multiforme (GBM) tumor genome from the Cancer Genome Atlas (TCGA) project and identified hundreds of CNVs, some were as small as 10 bp. We compared these CNAs with those detected using the array Comparative Genomic Hybridization (CGH) platforms and found that about one third were 'missed' by the array-CGH platforms, most of which were CNAs less than 10 kb. We selected 17 of the CNAs not detected by the array-based platforms for validation, ranging from 110 bp to 14 kb, and found that 15 of them are true CNAs. We further extended BIC-seq to the multi-sample case to identify recurrent CNAs in across multiple tumor genomes.