Non-targeted mass spectrometry (MS) has become an important method over recent years in the fields of metabolomics and environmental research. While more and more algorithms and workflows become available to process a large number of non-targeted data sets, there still exist few manually evaluated universal test data sets for refining and evaluating these methods. The first step of non-targeted screening, peak detection and refinement of it is arguably the most important step for non-targeted screening. However, the absence of a model data set makes it harder for researchers to evaluate peak detection methods. In this Data Descriptor, we provide a manually checked data set consisting of 255,000 EICs (5000 peaks randomly sampled from across 51 samples) for the evaluation on peak detection and gap-filling algorithms. The data set was created from a previous real-world study, of which a subset was used to extract and manually classify ion chromatograms by three mass spectrometry experts. The data set consists of the converted mass spectrometry files, intermediate processing files and the central file containing a table with all important information for the classified peaks.