Abstract Background Microarray analysis of immunoprecipitated chromatin (ChIP-chip) has evolved from a novel technique to a standard approach for the systematic study of protein-DNA interactions. In ChIP-chip, sites of protein-DNA interactions are identified by signals from the hybridization of selected DNA to tiled oligomers and are graphically represented as peaks. Most existing methods were designed for the identification of relatively sparse peaks, in the presence of replicates. Results We propose a data normalization method and a statistical method for peak identification from ChIP-chip data based on a mixture model approach. In contrast to many existing methods, including methods that also employ mixture model approaches, our method is more flexible by imposing less restrictive assumptions and allowing a relatively large proportion of peak regions. In addition, our method does not require experimental replicates and is computationally efficient. We compared the performance of our method with several representative existing methods on three datasets, including a spike-in dataset. These comparisons demonstrate that our approach is more robust and has comparable or higher power than the other methods, especially in the context of abundant peak regions. Conclusion Our data normalization and peak detection methods have improved performance to detect peak regions in ChIP-chip data.