Science education research has revealed a number of student mental models for atomic structure, among which, the one based on Bohr's model seems to be the most dominant. The aim of the current study is to investigate the coherence of these models when students apply them for the explanation of a variety of situations. For this purpose, a set of six tasks describing different everyday situations was given to 225 students of the 10th and 11th grades of secondary schools from Northern Greece. Quantitative analysis of the students' responses using Latent Class Analysis (LCA) showed that there is no consistency between models across the tasks and that the context of the task affects the distribution of students' responses across models. Qualitative analysis showed a variety of pieces of knowledge from different models that students combine when manipulating the tasks, which possibly causes a lack of consistency within each one of the models. The findings are discussed in terms of between and within model consistency, and the conclusions contribute to the debate concerning the coherent vs. fragmented knowledge hypotheses. The empirical evidence provided by the analysis clearly demonstrates that student mental models for atomic structure were not coherent when applied in different everyday situations. Implications for theory and practice are discussed.