In this research project we investigate the role of responses to anomalous data during modelling processes. Modelling is seen as a comprehensive practice that encompasses various aspects of scientific thinking; hence, it is an important style of scientific thinking, especially if analysed from a process-based perspective. Therefore, it provides the opportunity to understand the role of anomalous data on scientific thinking from a broader perspective. We analysed how pre-service biology teachers (N = 11) reacted to self-generated anomalous data during modelling processes induced by investigating a water black box. The videotaped and transcribed modelling processes were analysed using qualitative content analysis. If anomalous data were recognised, a majority of explanations were based on methodical issues. This finding supports results from previous studies investigating responses to first-hand anomalous data. Furthermore, we found four response patterns to anomalous data during modelling processes: "no recognition," "no explanation," "methodical explanation," and "model-related explanation." Besides, our study indicates by trend a systematic relation between response patterns to anomalous data and modelling strategies. Consequently, the improvement of responses to anomalous data could be a promising way to foster modelling competencies. We are convinced that an integrated approach to anomalous data and modelling could lead to deeper insights into the role of data in scientific thinking processes.