We propose a two-step normalization procedure for high-throughput mass spectrometry (MS) data, which is a necessary step in biomarker clustering or classification. First, a global normalization step is used to remove sources of systematic variation between MS profiles due to, for instance, varying amounts of sample degradation over time. A probability model is then used to investigate the intensity-dependent missing events and provides possible substitutions for the missing values. We illustrate the performance of the method with a LC-MS data set of synthetic protein mixtures.